2005-[sachdev s. sidhu] phage display in biotechnology

Boca Raton London New York Singapore

A CRC title, part of the Taylor & Francis imprint, a member of theTaylor & Francis Group, the academic division of T&F Informa plc.

Edited by

Sachdev S. Sidhu

Phage Display inBiotechnology andDrug Discovery

A phage-derived synthetic antibody against human death receptor DR5. At the top, a phage-displayed antigen-binding fragment (Fab) was used as a framework to present synthetic CDR loops derived from a binary codethat encodes only tyrosine and serine. At the center, synthetic Fab that recognizes human DR5 (red) with highaffinity and specificity was selected and the X-ray crystal structure was determined (PDB ID code 1ZA3). Atthe bottom, the structure reveals that the third complementarity determining region (CDR) of the heavy chainplays a dominant role in antigen recognition. The CDR loop contains a biphasic helix with tyrosine and serineresidues clustered on opposite faces, and the tyrosine face mediates contact with the antigen. The cover wasdesigned by Frederic Fellouse and David Wood, and structures were rendered with PyMOL (DeLano Scientific,San Carlos, CA).

Published in 2005 byCRC PressTaylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300Boca Raton, FL 33487-2742

© 2005 by Taylor & Francis Group, LLCCRC Press is an imprint of Taylor & Francis Group

No claim to original U.S. Government worksPrinted in the United States of America on acid-free paper10 9 8 7 6 5 4 3 2 1

International Standard Book Number-10: 0-8247-5466-2 (Hardcover) International Standard Book Number-13: 978-0-8247-5466-2 (Hardcover)

This book contains information obtained from authentic and highly regarded sources. Reprinted material isquoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable effortshave been made to publish reliable data and information, but the author and the publisher cannot assumeresponsibility for the validity of all materials or for the consequences of their use.

No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic,mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, andrecording, or in any information storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.com(http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive,Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registrationfor a variety of users. For organizations that have been granted a photocopy license by the CCC, a separatesystem of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used onlyfor identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Catalog record is available from the Library of Congress

Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com

and the CRC Press Web site at http://www.crcpress.com

Taylor & Francis Group is the Academic Division of T&F Informa plc.

DK2163_Discl.fm Page 1 Wednesday, April 13, 2005 8:12 AM

Preface

To my parents and Sabrina,for their support.

Foreword

Science has always progressed by coupling insightful observa-tions leading to testable hypotheses with innovative technol-ogies that facilitate our ability to observe and test them. Inthe field of protein science the technologies for protein displayand in vitro selection have had an enormous impact on ourability to probe and manipulate protein functional properties.

The development of site-directed mutagenesis, whichallowed one to systematically probe a gene sequence in thelate 1970s, gave birth to the field of protein engineering inthe early 1980s. Throughout the 1980s most scientists inthe protein engineering field would generate and purify onemutant protein at a time and characterize its functional pro-perties. Some investigators had developed selections andscreens that allowed one to test many variants simulta-neously, but these tended to be highly specific for certain pro-teins (notably DNA binding proteins) and focused primarilyon studying protein stability. Moreover, the selections weregenerally done in the context of a living cell, which limitedthe range of assays that could be performed. While replicaplating screens were available to test variant proteins out of

v

the cell, these tended to be quite labor intensive, thus limitingthe number of variants that could be screened.

In 1985, George Smith published a paper showing thatsmall peptides derived from EcoRI could be inserted intothe gene III attachment protein in filamentous bacterialphage, which could then be captured using antibodies to thesmall peptide. This observation incubated several years andthen, in the late 1980s and early 1990s, other groups showedit was possible to display whole proteins on gene III that werefolded and capable of binding their cognate ligands. Moreoverit was shown that by appropriate manipulation of copy num-ber on the phage it was possible to select a range of bindingaffinities, from weak at high copy number to strong at lowcopy number. These selections could all be done in vitro andunder a variety of selection conditions, limited only by bindingto a support-bound ligand.

Throughout the 1990s up to today, huge improvementshave been made to the display technology allowing massiveincreases in library number (now routinely > 1010 variantsper selection), recursive mutagenesis cycles allowing one tomutate as one selects, new display formats including otherphage species, bacteria, yeast and ribosomes, and automationto further simplify the process. As with any technology thereare limitations. For example, not all proteins can be readilydisplayed on phage and expression effects can bias the out-come of the selection. Nonetheless, phage display has had ahuge impact on probing, improving and designing new func-tional properties into proteins and peptides including bindingaffinity, selectivity, catalysis, chemical and thermal stabilityamong others. This book edited by Sachdev Sidhu providesan excellent review of the state-of-the-art in phage displaytechnology now and in the near future.

James A. WellsPresident and Chief Scientific Officer

Sunesis PharmaceuticalsSouth San Francisco, California, U.S.A.

vi Foreword

Preface

Recent years have witnessed the sequencing of numerousgenomes, including the all-important human genome itself.While genomic information offers considerable promise fordrug discovery efforts, it must be remembered that we livein a protein world. The vast majority of biological processesare driven by proteins, and the full benefits of DNA databaseswill only be realized by the translation of genomic informationinto knowledge of protein function. Ultimately, drug discoverydepends on themanipulation andmodification of proteins, andthus, the genomic panacea comes with significant challengesfor life scientists in the field of therapeutic biotechnology.Indeed, it has become clear that success in the modern era ofbiology will go to those who apply to protein analysis thehigh-throughput principles that made whole-genome sequen-cing a reality.

In this context, phage display is an established combina-torial technology that is likely to play an even greater role inthe future of drug discovery. The power of the technologyresides in its simplicity. Rapid molecular biology methodscan be used to create vast libraries of proteins displayed on

vii

bacteriophage that also encaspulate the encoding DNA.Billions of different proteins can be screened en masse andindividual protein sequences can be decoded rapidly fromthe cognate gene. In essence, the technology enables theengineering of proteins with simple molecular biology techni-ques that would otherwise only be applicable to DNA. Inaddition, the technology is very much suited to the methodscurrently used for high-throughput screening, and thus, canbe readily adapted to the analysis of multiple targets andpathways.

This book comprises 17 chapters that provide a compre-hensive view of the impact and promise of phage display indrug discovery and biotechnology. The chapters detail thetheories, principles, and methods current in the field anddemonstrate applications for peptide phage display, proteinphage display, and the development of novel antibodies. Thebook as a whole is intended to give the reader an overviewof the amazing breadth of the impact that phage display tech-nology has had on the study of proteins in general and thedevelopment of protein therapeutics in particular. I hope thatthis work will serve as a comprehensive reference forresearchers in the phage field and, perhaps more importantly,will serve to inspire newcomers to adapt the technology totheir own needs in the ever expanding world of therapeuticbiology.

Sachdev S. Sidhu

viii Preface

Contents

Foreword James A. Wells . . . . vPreface . . . . viiContributors . . . . xv

1. Filamentous Bacteriophage Structure andBiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1Diane J. Rodi, Suneeta Mandava, and Lee Makowski

I. Introduction . . . . 1II. Taxonomy and Genetics . . . . 3III. Viral Gene Products . . . . 5IV. Structure of the Virion . . . . 11V. Filamentous Bacteriophage Life Cycle . . . . 18VI. Phage Library Diversity . . . . 34VII. Biological Bottlenecks: Sources of Library

Censorship . . . . 35VIII. Quantitative Diversity Estimation . . . . 41IX. Improved Library Construction . . . . 45

References . . . . 47

ix

2. Vectors and Modes of Display . . . . . . . . . . . . . . 63Valery A. Petrenko and George P. Smith

I. Introduction . . . . 63II. Most Display Vectors are Based on Filamentous

Phage . . . . 65III. General Cloning Vectors Based on Filamentous

Phage . . . . 71IV. Classification of Filamentous Phage Display

Systems . . . . 75V. Phage f1—The First Phage-Display Vector . . . . 77VI. Low DNA Copy Number Display Vectors Based

on fd-tet . . . . 78VII. Diversity of Type 3 Vectors . . . . 80VIII. Type 8 Vectors: First Lessons . . . . 81IX. Mosaic Display in Type nn Systems . . . . 83X. Mosaic Display in Phagemid Systems . . . . 89XI. Vectors for C-Terminal Display . . . . 91XII. Phage Proteins as Constraining

Scaffolds . . . . 93XIII. Conclusion . . . . 95


3. Methods for the Construction of Phage-DisplayedLibraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111Frederic A. Fellouse and Gabor Pal

I. Introduction . . . . 111II. Oligonucleotide-Directed Mutagenesis . . . . 112III. Random Mutagenesis . . . . 123IV. Combinatorial Infection and Recombination . . . . 126V. DNA Shuffling . . . . 129


4. Selection and Screening Strategies . . . . . . . . . . 143Mark S. Dennis

I. Introduction . . . . 143II. General Considerations . . . . 144III. The Selection Process . . . . 146IV. Selections Methods . . . . 150


x Contents

5. Phage Libraries for Developing Antibody-TargetedDiagnostics and Vaccines . . . . . . . . . . . . . . . . . . 165Nienke E. van Houten and Jamie K. Scott

I. Introduction . . . . 165II. Phage-Display Libraries as Tools for Epitope

Discovery . . . . 170III. Diagnostics . . . . 179IV. Phage Libraries for Epitope Mapping . . . . 187V. Phage Display Libraries for Vaccine

Development . . . . 200VI. Developing Immunogens from Peptide

Leads . . . . 218VII. Summary . . . . 235VIII. Conclusion . . . . 238IX. Abbreviations . . . . 239


6. Exploring Protein–Protein Interactions UsingPeptide Libraries Displayed on Phage . . . . . . . 255Kurt Deshayes

I. Introduction . . . . 255II. Extracellular Protein–Protein Interactions . . . . 256III. Intracellular Protein–Protein Interactions . . . . 268IV. Conclusions . . . . 274


7. Substrate Phage Display . . . . . . . . . . . . . . . . . . 283Shuichi Ohkubo

I. Overview . . . . 283II. Introduction . . . . 284III. The Concept of Substrate Phage Display . . . . 285IV. Application of Substrate Phage Display to Cancer

Research . . . . 294V. Conclusions . . . . 305


8. Mapping Intracellular Protein Networks . . . . . 321Zhaozhong Han, Ece Karatan, and Brian K. Kay

I. Introduction . . . . 321II. Domain-Mediated Interactions . . . . 323

Contents xi

III. Nondomain Mediated Protein–ProteinInteractions . . . . 336

IV. Software for Identifying Candidate InteractingPartners . . . . 336

V. Analyzing Predicted Interactions . . . . 337VI. Relevance to Biotechnology and Drug

Discovery . . . . 338References . . . . 340

9. High Throughput and High Content ScreeningUsing Peptides . . . . . . . . . . . . . . . . . . . . . . . . . . . 347Robert O. Carlson, Robin Hyde-DeRuyscher, andPaul T. Hamilton

I. Introduction . . . . 347II. Peptides as Enzyme Inhibitors . . . . 348III. Peptides as Conformational Probes . . . . 355IV. Summary . . . . 376


10. Engineering Protein Foldingand Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . 385Mihriban Tuna and Derek N. Woolfson

I. Protein Redesign and Design . . . . 385II. Early Combinatorial Studies Aimed at Repacking

the Cores of Proteins . . . . 387III. Phage Display in Engineering Protein

Stability . . . . 390IV. A Worked Example: Repacking the Hydrophobic

Core of Ubiquitin . . . . 397V. Studies that Build on the Original Methods . . . . 406VI. Summary . . . . 408


11. Identification of Natural Protein–ProteinInteractions with cDNA Libraries . . . . . . . . . . 415Reto Crameri, Claudio Rhyner, Michael Weichel,Sabine Fluckiger, and Zoltan Konthur

I. Overview . . . . 415II. Introduction . . . . 416III. Cloning Vectors . . . . 417

xii Contents

IV. Display of cDNA Libraries on Phage Surface . . . . 420V. Problems Associated with the Display of cDNA

Libraries on Phage Surface . . . . 425VI. Adaptability of Phage Display to High-Throughput

Screening Technology . . . . 427VII. Conclusions . . . . 428


12. Mapping Protein Functional Epitopes . . . . . . 441Sara K. Avrantinis and Gregory A. Weiss

I. Introduction . . . . 441II. Single Point Alanine Mutagenesis . . . . 443III. Combinatorial Site-Specific Mutagenesis . . . . 447IV. Other Approaches to Phage-Displayed Functional

Epitope Mapping . . . . 455V. Conclusion . . . . 456


13. Selections for Enzymatic Catalysts . . . . . . . . . 461Julian Bertschinger, Christian Heinis, and Dario Neri

I. Introduction . . . . 461II. Selection Methods . . . . 464III. Discussion . . . . 482


14. Antibody Humanization and Affinity MaturationUsing Phage Display . . . . . . . . . . . . . . . . . . . . . 493Jonathan S. Marvin and Henry B. Lowman

I. Introduction . . . . 493II. Humanization Using Phage Display . . . . 497III. In Vitro Affinity Maturation of Antibodies . . . . 501IV. Emerging Approaches . . . . 519V. Conclusions . . . . 520


15. Antibody Libraries from ImmunizedRepertoires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529Jody D. Berry and Mikhail Popkov

I. Introduction . . . . 529II. Immune Antibody Library Construction . . . . 538

Contents xiii

III. Immune Antibody Library Selection . . . . 570IV. The Future . . . . 622


16. Naıve Antibody Libraries from NaturalRepertoires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659Claire L. Dobson, Ralph R. Minter, andCelia P. Hart-Shorrock

I. Introduction . . . . 659II. Construction of Naıve Libraries . . . . 660III. Applications of Naıve Libraries . . . . 674IV. Summary . . . . 700


17. Synthetic Antibody Libraries . . . . . . . . . . . . . . 709Frederic A. Fellouse and Sachdev S. Sidhu

I. Introduction . . . . 709II. The Scripps Research Institute . . . . 711III. The Medical Research Council . . . . 714IV. Morphosys . . . . 720V. Genentech . . . . 726VI. Conclusions . . . . 732


Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 741

xiv Contents

Contributors

Sara K. Avrantinis Department of Chemistry, University ofCalifornia, Irvine, California, U.S.A.

Jody D. Berry Monoclonal Antibody Section, ReagentDevelopment Unit, National Centre for Foreign Animal Disease,Winnipeg, Manitoba, Canada

Julian Bertschinger Institute of Pharmaceutical Sciences,Swiss Federal Institute of Technology, Zurich, Switzerland

Robert O. Carlson Karo Bio USA, Inc., Durham, NorthCarolina, U.S.A.

Reto Crameri Swiss Institute of Allergy and Asthma Research(SIAF), Davos, Switzerland

Mark S. Dennis Department of Protein Engineering, Genentech,Inc., South San Francisco, California, U.S.A.

xv

Kurt Deshayes Department of Protein Engineering, Genentech,Inc., South San Francisco, California, U.S.A.

Claire L. Dobson Cambridge Antibody Technology, Cambridge,U.K.

Frederic A. Fellouse Department of Protein Engineering,Genentech, Inc., South San Francisco, California, U.S.A.

Sabine Fluckiger BioVision Schweiz AG, Davos, Switzerland

Paul T. Hamilton Karo Bio USA, Inc., Durham, North Carolina,U.S.A.

Zhaozhong Han Argonne National Laboratory, BiosciencesDivision, Argonne, Illinois, U.S.A.

Celia P. Hart-Shorrock Cambridge Antibody Technology,Cambridge, U.K.

Christian Heinis Institute of Pharmaceutical Sciences, SwissFederal Institute of Technology, Zurich, Switzerland

Robin Hyde-DeRuyscher Karo Bio USA, Inc., Durham, NorthCarolina, U.S.A.

Ece Karatan Argonne National Laboratory, Biosciences Division,Argonne, Illinois, U.S.A.

Brian K. Kay Argonne National Laboratory, BiosciencesDivision, Argonne, Illinois, U.S.A.

Zoltan Konthur Max Planck Institute of Molecular Genetics,Berlin, Germany

Henry B. Lowman Department of Antibody Engineering,Genentech, Inc., South San Francisco, California, U.S.A.

Lee Makowski Combinatorial Biology Unit, BiosciencesDivision, Argonne National Laboratory, Argonne, Illinois, U.S.A.

xvi Contributors

Suneeta Mandava Combinatorial Biology Unit, BiosciencesDivision, Argonne National Laboratory, Argonne, Illinois, U.S.A.

Jonathan S. Marvin Department of Antibody Engineering,Genentech, Inc., South San Francisco, California, U.S.A.

Ralph R. Minter Cambridge Antibody Technology, Cambridge,U.K.

Dario Neri Institute of Pharmaceutical Sciences, Swiss FederalInstitute of Technology, Zurich, Switzerland

Shuichi Ohkubo Cancer Research Laboratory, Hanno ResearchCenter, TAIHO Pharmaceutical Co., Ltd., Hanno, Saitama, Japan

Gabor Pal Department of Biochemistry, Eotvos LorandUniversity, Budapest, Hungary

Valery A. Petrenko Department of Pathobiology, College ofVeterinary Medicine, Auburn University, Auburn, Alabama, U.S.A.

Mikhail Popkov Department of Molecular Biology, The ScrippsResearch Institute, La Jolla, California, U.S.A.

Claudio Rhyner Swiss Institute of Allergy and Asthma Research(SIAF), Davos, Switzerland

Diane J. Rodi Combinatorial Biology Unit, Biosciences Division,Argonne National Laboratory, Argonne, Illinois, U.S.A.

Jamie K. Scott Department of Molecular Biology andBiochemistry, Simon Fraser University, Burnaby, BritishColumbia, Canada

Sachdev S. Sidhu Department of Protein Engineering,Genentech, Inc., South San Francisco, California, U.S.A.

George P. Smith Division of Biological Sciences, University ofMissouri, Columbia, Missouri, U.S.A.

Mihriban Tuna Department of Biochemistry, School of LifeSciences, University of Sussex, Falmer, U.K.

Contributors xvii

Nienke E. van Houten Department of Molecular Biology andBiochemistry, Simon Fraser University, Burnaby, BritishColumbia, Canada

Michael Weichel Swiss Institute of Allergy and AsthmaResearch (SIAF), Davos, Switzerland

Gregory A. Weiss Department of Chemistry, University ofCalifornia, Irvine, California, U.S.A.

Derek N. Woolfson Department of Biochemistry, School of LifeSciences, University of Sussex, Falmer, U.K.

xviii Contributors

1

Filamentous BacteriophageStructure and Biology

DIANE J. RODI, SUNEETA MANDAVA, andLEE MAKOWSKI

Combinatorial Biology Unit, BiosciencesDivision, Argonne National Laboratory,

Argonne, Illinois, U.S.A.

I. INTRODUCTION

Phage display technology provides a remarkably versatile toolfor exploring the interactions between proteins, peptides, andsmall molecule ligands. As such it has become widely adaptedfor use in epitope mapping, identification of protein–peptideand protein–protein interactions, protein–small moleculeinteractions, humanization of antibodies, identification oftissue-targeting peptides, and many other applications, asoutlined throughout this book. However, it must be kept inmind that phage display is a combinatorial biology approach,

1

not a combinatorial chemistry approach. The great strength ofphage display over combinatorial methods that are strictlychemical is that the isolation of a single interacting proteinor peptide attached to a phage particle is sufficient to allowthe complete characterization of the isolate: the interactingvirus can be grown up in bulk and the sequence of the dis-played protein or peptide inferred from the DNA sequencecarried within the viral particle. The other side of this coinis that phage display technology utilizes living systems, andis therefore constrained in its potential diversity by the mole-cular requirements of those systems.

The biological limitations that impact phage displaytechnology are defined not simply by viral structure, but bythe well-balanced phage–host system as a whole. The displayof a protein or peptide on the surface of a bacteriophage par-ticle involves insertion of the corresponding DNA into thegene of a structural protein and the expression of the foreignsequence as a fusion with the structural protein in such a waythat it is exposed, at least in part, on the surface of the phageparticle. This process perturbs the phage–host system andmay result in anything from a negligibly small alteration inphage growth rate to a complete halt of phage production.Disruption of any step along the way between DNA cloningand production of virus, including protein synthesis, proteintranslocation, viral morphogenesis, viral stability, host cellbinding or subsequent steps in the infection process, canremove a particular display construct from the final phagepopulation. Additionally, in the context of library screeningmethodology, it is also important to note that different insertsplaced at the same site may have very different effects on therate of viral production, resulting in biases that can seriouslyimpact the diversity of a phage-displayed library and, conse-quently, the results of affinity selection experiments. Somemembers of the libraries are present at much lower levelsthan others, whereas others are absent. These biases mustbe well characterized in order to make optimal use of librariesin affinity selections or other experiments designed to takeadvantage of the unique properties of display libraries. There-fore, in order to understand the effect of biology on phage

2 Rodi et al.

display, the way the phage interacts with its host must beconsidered in detail.

In this chapter, we outline the steps of phage–host inter-action and discuss how those interactions may impact thediversity of phage-displayed libraries. Understanding theselimitations in more detail should provide a starting point forengineering methods to minimize their effect on the use ofphage display technology within the broad range of applica-tions reviewed in this volume. Except for DNA replication,each step appears to have a detectable effect on the expressionof some members of some libraries. Some effects appear sig-nificant, whereas others are barely detectable. At the end,we briefly review identified bottlenecks in the viral life cycleand suggest simple strategies that can be implemented forminimizing the perturbations.

II. TAXONOMY AND GENETICS

The filamentous bacteriophages are a family of ssDNA-containing viruses (genus Inovirus) that infect a widevariety of gram-negative bacteria, including Escherichia coli,Xanthomonas, Thermus, Pseudomonas, Salmonella andVibrio. The best characterized of the filamentous phage arethe Ff class of viruses, so named because of their method ofhost cell entry via the tip of the F conjugative pilus on the sur-face of male E. coli cells. The Ff viruses include M13, fd, andf1, all of which possess a 98% identity at the DNA sequencelevel. Ff virus particles are long, slender, and flexible rods,with a diameter of about 65 A. The wild-type Ff phages arebetween 0.8 and 0.9mm long, giving the virus the proportionsof a 4 foot long pencil. Various engineered strains have some-what longer genomes, with the length of the particleincreased proportionate to the length of the encapsulatedDNA. Although there is considerable heterogeneity withinthe family, some similarities of sequence and genome organi-zation are discernable among all group members. An electronmicrograph shown in Fig. 1 gives a rough idea of the propor-tions of the phage particles. The single-stranded, circular

Filamentous Bacteriophage Structure and Biology 3

genome occupies the axis of the particle, stretched out foralmost the entire length of the virion. Virus lengths aredependent upon both the size of the enclosed genome as sta-ted above and on the physical distribution of the DNA withinthe capsid (axial distance per base), the latter of which hasbeen demonstrated to be major coat protein charge dependent(1). Little substructure is visible except at the end involved inhost cell attachment. Each crosssection of the virion has an‘‘up’’ strand and a ‘‘down’’ strand present, but these are notbase paired since there is no complimentary relationshipbetween the sequences of the two strands except within thehairpin which acts as the packaging signal that nucleatesthe initiation of viral assembly.

Figure 1 Electron micrograph of bacteriophage M13. This micro-graph of M13 phage particles visually demonstrates the rationalefor their designation as ‘‘filamentous’’ bacteriophage. The aminoterminus of at least four copies of the gene III protein are visibleat the end of the phage particle; the two subtilisin-cleavable N1and N2 domains are seen as knobby structures at the proximalend of the virus. (Micrograph courtesy of Irene Davidovich.)

4 Rodi et al.

Ff viruses are not lytic, but rather parasitic. Productiveinfections result in viral release via extrusion or secretionacross the inner and outer bacterial membranes in theabsence of host cell death, with the infected cells continuingto grow and divide (albeit at a significantly reduced rate).M13 produces anywhere from 200 to 2000 progeny phageper cell per doubling time (2,3). This phage production repre-sents a serious metabolic load for the infected E. coli, withphage proteins making up 1–5% of total protein synthesisand resulting in a reduction in cell growth of 30–50% fromuninfected cells. The nonlytic nature of Ff infection, alongwith the simultaneous presence of both single- and double-stranded forms of viral DNA, little size constraint on insertedDNA, and an exceptionally high viral titer capacity (typically1011–1012 particles per mL), has made the filamentous phage,primarily M13, a workhorse for molecular biology for the last20 years.

III. VIRAL GENE PRODUCTS

The Ff phage genome encodes a total of 11 proteins (see Fig. 2for genome organization). There are five structural proteins,all of which are inserted into the inner host cell membraneprior to assembly (see Fig. 3 for overall structural organiza-tion of Ff phage). pVIII and pIII are synthesized with signalsequences that are removed subsequent to membrane inser-tion, whereas pVI, pVII, and pIX are absent signal sequences.Three nonstructural phage proteins, pI, pXI, and pIV arerequired for phage morphogenesis, but are not incorporatedinto the phage structure. pX and pXI are the result of in-frame internal translation initiation events in genes II andI, respectively, and are identical with the C-terminal portionsof pII and pI in amino acid sequence, membrane localizationand topology (4). In addition to the coding regions, there isan intergenic region which contains the signals for the initia-tion of synthesis of both the plus (þ) or viral-contained DNAstrand, and the minus (–) strand; the initiation of capsidassembly signal (or packaging signal, PS) which lies between


the (–) origin and the end of the pIV gene; and the signal fortermination of RNA synthesis. Parts of the intergenic regionhave been shown to be dispensable (reviewed in Ref. 3), butall of the coding region products are necessary for the synth-esis of infectious progeny phage.

III.A. Replication Proteins (pII and pX)

pII is a 410 amino acid protein (MW¼ 46,137) which isrequired for all phage-specific DNA synthesis other than theformation of the complementary strand of the infecting ssDNAby host enzymes. pII has both endonuclease and topoisomeraseactivities required during the DNA replication phase of infec-tion. pX is a 111 residue protein (MW¼ 12,672) which isencoded entirely within gene II, initiating at codon 300, anAUG that is in phase with the initiating AUG of gene II.

Figure 2 Ff phage genome. The location of each viral gene is indi-cated by number, with the direction of transcription shown byarrow. The origin of replication lies within the intergenic region,between the genes for pIV and pII. The packaging signal (PS) liesbetween the (–) strand origin of replication and gene IV.

6 Rodi et al.

Although pX has the same amino acid sequence as the car-boxyl-terminal end of pII, it has been shown to possess uniquefunctions within the viral life cycle, such as inhibition of pIIfunction (5).

Figure 3 Schematic diagram of the Ff bacteriophage. This dia-gram depicts the structural organization of M13 as a representativeof the Ff viral family. At the top of the diagram lies the distal end ofthe particle, at which viral assembly initiates. At the bottom of thefigure is the proximal or infectious end of the virus, with five copiesof the pIII anchored to the particle by five copies of the pVI.


III.B. Single-Stranded DNA Binding Protein (pV)

Gene V codes for an 87 amino acid protein (MW¼ 9682) thatexists as a stable dimer even at a concentration as low as1nM (6). The crystal structure of the protein has been solvedto 1.8 A resolution using multiwavelength anomalous disper-sion on a selenomethionine-containing protein, and is shownin Fig. 4 (7). Each monomer is largely b-structure, with 58

Figure 4 Crystal structure of the gene V ssDNA-binding protein.The crystal structure of pV has been solved to 1.8 A resolution usingmultiwavelength anomalous dispersion on a selenomethionine deri-vative and is shown here in a backbone format (7) (PDB accessioncode 2GN5). The protein normally exists as a dimer and wrapsaround the single-stranded form of the viral DNA within the hostcell cytoplasm. Residues Tyr26, Leu28, and Phe73 have been shownto be critical for DNA binding (7) (residues shown in spacefillformat).

8 Rodi et al.

out of 87 residues arranged in a five-stranded antiparallelb-sheet; two antiparallel b-ladder loops protrude from thissheet. The remainder of the molecule is arranged into 310-helices (residues 7–11 and 65–67), b-bends (residues 21–24,50–53, and 71–74) and one five-residue loop (residues 38–42).Nuclear magnetic resonance (NMR) analysis of the gene V pro-tein (8) suggests that the DNA binding loop (residues 16–28;see Fig. 4) is flexible in solution. This protein serves the dualfunctions of sequestering the intracellular ssDNA viral gen-omes (3) and modulating the translation of gene II mRNA (9).

III.C. Major Structural Protein (pVIII)

Gene VIII codes for the major coat protein of the virus. Themajor coat proteins of all filamentous phages are short, ran-ging from 44 to 55 amino acids, with most being encoded witha signal sequence (Pseudomonas aeruginosa-infecting phagePf3, being an example of a pVIII absent a signal sequence).In the Ff group of phage, the major coat protein is 50 aminoacids long (MW¼ 5235), with a 23 amino acid long signalsequence. Approximately 2800 copies of pVIII are requiredto coat one full-length wild-type Ff virion. The concentrationof pVIII in the inner cell membrane is very high—at least5� 105 molecules of pVIII are exported as virions per infectedcell per doubling, making it one of the most abundant proteinsin the infected cell (10).

III.D. Minor Structural Proteins (pIII, pVI, pVII,and pIX)

Each end of the filamentous phage particle is distinctive, bothin function and in protein composition. The distal end of thephage contains approximately five copies each of the twosmall hydrophobic proteins, pVII (33 a.a., MW¼ 3,599) andpIX (32 a.a., MW¼ 3,650). In the absence of either pVII orpIX, almost no phage particles are formed (11), and geneticevidence suggests that both are involved in initiation ofassembly (12).

Genes III (coding for pIII, 406 a.a., MW¼ 42,522) and VI(coding for pVI, 112 a.a., MW¼ 12,342) encode two minor coat


proteins which sit at the end of the viral particle thatextrudes last during assembly and is also responsible for hostcell binding. pIII serves dual functions—it is required forinfectivity, and it is necessary for termination of viral assem-bly. In the absence of pIII, noninfectious multiple-length‘‘polyphage’’ particles are produced, which contain severalunit-length genomes. The crystal structures of the twoamino-terminal domains (D1 or N1 and D2 or N2) of pIII havebeen solved for both phages fd (13) and M13 (14) (see Fig. 5).These structures demonstrate that although the individualdomains are the same, they differ by a rigid body rotation of

Figure 5 Crystal structure of the two amino-terminal domains ofthe gene III protein. The crystal structure of N1 and N2 at theamino-terminus of pIII has been solved to 1.46 A resolution usingmultiwavelength anomalous dispersion on a selenomethionine deri-vative and is shown here in cartoon format with the N1 domainhighlighted in dark blue (14) (PDB accession code 1G3P). Thesetwo domains can be seen as knobs at the proximal end of M13 asseen in the micrograph in Fig. 1.

10 Rodi et al.

N2 with respect to N1 around a hinge located at the end of theshort antiparallel b-sheet connecting N1 and N2.

III.E. Morphogenetic Proteins (pI, pIV, and pXI)

Proteins pI (348 a.a.,MW¼ 39,502), pIV (405 a.a.,MW¼ 43,476)and pXI (108 a.a.,MW¼ 12,424) are required for viral assembly,but arenotpresent in thefinal intact virus particle (for a review,seeRef. 15). pI spans the innermembrane and functions inmul-tiple ways during phage assembly (see Sec. V), including inter-acting with both host cell factors and phage proteins requiredfor morphogenesis, and in helping in the formation of phage-specific adhesion zones. pIV is an integral outermembrane pro-tein whose carboxyl-terminal half mediates formation of a mul-timer that appears to be composed of between 10 and 12 pIVmonomers (16,17). The morphogenetic proteins pI and pIV,located in different membranes, appear to interact to form anexit structure through which the assembling viral particleextrudes from the host cell (18). pXI is the result of an internaltranslational initiation within gene I, is more abundant ininfected cells than pI, and is believed to be an essential part ofa ‘‘preassembly complex’’ consisting of pI, pIV, and pXI (19).

IV. STRUCTURE OF THE VIRION

IV.A. Overall Structural Organization

Two symmetry classes of filamentous phage have been identi-fied by X-ray diffraction: Class I, which includes fd, M13, f1,and Ike, and Class II, which includes Pf1 and Xf (20). Figure1 is an electron micrograph of M13, and Fig. 3 is a schematicof the organization and relative positions of its protein com–ponents. The coat proteins from the two classes have lengthsthat vary from 44 to 53 amino acids and sequences with simi-lar overall character (each has an amino-terminal end rich inacidic residues, a central hydrophobic region, and a carboxyterminus rich in basic residues) but little conserved sequence.A large fraction of the viral mass (around 85%) is made up ofmany copies of pVIII, which forms a 15–20 A thick flexiblecylinder about the single-stranded viral genome. There are


approximately 0.435 copies of pVIII per nucleotide (21), andthe length of the virion is about 3.3 A per pVIII (or 1.435 Aper nucleotide) plus approximately 175 A of minor coat pro-teins (22). The amino terminus of pVIII is exposed to the sur-face, with the first four to five residues forming a flexible armthat appears to extend away from the virus. The distal or topend of the particle is assembled first, and contains approxi-mately three to four copies each of pVII and pIX asdetermined by labeling studies (23), which may form a hydro-phobic plug at the top of the virus (24). The proximal end(bottom in Fig. 3) contains approximately five copies each ofpVI, which is believed to mediate attachment of approxi-mately five copies of pIII to the body of the virus.

IV.B. pVIII Structure

In the intact M13 viral particle, pVIII molecules are arrangedwith a fivefold rotational axis and a twofold screw axis, with apitch of about 32 A (25). Most of the pVIII structure appearsto be comprised of a gently curving a-helix extending fromPro6 to near the carboxyl-terminus (26–29). The axis of thepVIII a-helix is tilted about 20� relative to the virion axisand wraps around the virion axis in a right-handed helicalsense, as can be seen in Fig. 3. The amino acid sequence ofpVIII can be broken into four parts: a mobile surface segment(Ala1–Asp4 or Asp5); an amphipathic a-helix extending fromPro6 to about Tyr24; a highly hydrophobic helix extendingfrom Ala25 to Ala35; and an amphipathic helix forming theinside wall of the coat, extending from Thr36 to Ser50 (27).This last helical region contains four positively charged sidechains which interact with the phosphate backbone of theDNA, and the DNA bases face inward (28,30).

The amino-terminal half of pVIII is the region in whichalmost all fusions to pVIII have been constructed. The aminoterminus lies at the surface, with only the first three resi-dues accessible to digestion by proteases (31) and theremainder of the phage surface is composed largely of theamphipathic helix that extends from Pro6 to about Tyr24.Random peptides inserted at or near the amino terminus

12 Rodi et al.

of pVIII have shed some light on the packing requirementsand structure of the phage particle surface. Fiber diffractionanalysis of an insertion mutant, with a pentapeptide(GQASG) inserted between residues 4 and 5, indicated thatthe insert lies in an extended conformation within a shallowgroove between two adjacent a-helices on the viral surface(32). Given the resemblance of this arrangement to the pre-sentation of peptides by major histocompatibility antigens(33), the authors postulate that the three-dimensional con-formation of small pVIII inserts may contribute to the highimmunogenicity observed for peptides inserted into the geneVIII product of M13. The length of the shallow surfacegroove (17 to 20 A) corresponds to the upper length limitobserved when foreign peptides are fused to all 2800 copiesof pVIII (i.e., six to eight residues; see Refs. 34–36). Alterna-tive explanations for the size limitation include reduced sig-nal peptidase cleavage rates for some mutant procoatproteins (35,37), possible interference with pVIII=pVII inter-actions during assembly initiation (38) and=or the physicalimpossibility of extruding a virion of enlarged diameterthrough the 7nm exit pore formed by the pI=pXI=pIV com-plex (24,39).

The carboxyl terminus of pVIII is at low radius, roughly20 A from the surface of the virion in close proximity to theviral DNA. Recent work by Fuh et al. (40) has shown thatviable phage particles can be constructed which display, atvery low levels, foreign peptides fused to the carboxyl end ofpVIII. This display requires a minimum of 8–10 residues asa linker to render the carboxy-terminal fusion product acces-sible to the surface. These data appear to be consistent withtwo possible scenarios: either the linker extends throughthe protein coat and disrupts the local packing of pVIII mole-cules around the DNA core; or alternatively, the inserts areonly tolerated near the proximal end of the virion where thecarboxyl terminus of pVIII may be closer to the virus surface.This latter hypothesis is consistent with the fact (41) that dis-play with the carboxy-terminal pVIII fusion was reducedabout 10-fold relative to amino-terminal fusions on pIII,indicating recombinant pVIII levels of roughly 0.1–0.2 per


viral particle (given the roughly 1–2 fusion pIII moleculesregularly achieved per particle).

Filamentous phages are notoriously resistant to numer-ous physicochemical assaults, including prolonged incubationat high temperatures, in nonionic detergents, in high salt,and at low pH. However, Class I viral particles have beenshown to be sensitive to inactivation by small organics suchas chloroform, with accompanying structural collapse torod-shaped particles (42). Sequence analysis of isolatedchloroform-resistant and growth-enhanced mutants of pVIII(36,43,44) have revealed numerous point mutations whichmap throughout a single slice of the viral particle, with a par-ticular hot spot for mutation at the Pro6=Val31 location at thesurface. Modeling studies, based upon fiber diffraction analy-sis (27,28) indicate the presence of a substantial depression orhole in a large hydrophobic surface patch at this location,with Val31 lying exposed at the bottom (43). The V31L cloneisolated by Oh et al. (43) and a growth-enhanced recombinantpVIII, isolated by Iannolo et al. (44), that contained a tripep-tide (PFP) insertion may succeed by ‘‘plugging’’ this hydro-phobic hole which extends to the phage surface, resulting inan overall more stable structure. Studies on the stereochem-istry of this site suggest that a chloroform molecule canindeed fit into the Val31 hole (43). These data are consistentwith that of Weiss et al. (45), who isolated numerous pointmutants of pVIII with a P6F conversion which exhibitedgreater than 10-fold enhancement in the efficiency of displayof human growth hormone on the surface of M13, possibly theresult of stabilized phage particles.

IV.C. Distal End Structure

The distal end of the phage particle has approximately fivecopies each of the two small proteins pVII and pIX. Thisend contains the packaging sequence (PS) and is the part ofthe phage assembled first. The gene IX protein has beenshown to exist as 67% a-helix in lipid bilayers, with its posi-tively charged carboxyl terminus residing on the outsideof the membrane and thus available for binding to the

14 Rodi et al.

negatively charged DNA hairpin loop of the PS (46,47). Usingsera raised against pVII and pIX, Endemann and Model (38)have shown that pIX is accessible in the intact phage particle,whereas at least some epitopes of pVII are not. Immunopreci-pitation experiments on detergent-disrupted virus suggest aninteraction between pVII and pVIII in the virus particle. Gaoet al. (48) have shown that the amino termini of pVII and pIXcan be used to display the antibody heavy chain variabledomain (VH) and light chain variable domain (VL), respec-tively, and that this heterodimeric presentation affords aviable antibody variable fragment (Fv) with fully functionalbinding and catalytic activities. Single-chain Fv (scFv)libraries have been displayed as fusions with the amino termi-nus of pIX (49).

IV.D. Proximal End

The proximal, or infectious end, consists of approximately fivecopies each of pIII and pVI. Antibodies directed against pVIdo not interact with intact phage, suggesting that pVI issomewhat buried within the viral particle (38). A partialmodel for this end of the virus was postulated by assumingthat the amino-terminal half of pVI is structurally homolo-gous to two pVIII molecules (24) as shown in Fig. 3. At thisend of the virus, termination of assembly leaves two layersof pVIII exposed to solvent, as there are no continuing pVIIIrings to cover them. This potentially unstable situation couldbe ameliorated by a fold of pVI, much of which forms anamphipathic surface layer, up and around these last twopVIII rings, holding the virion together and helping to main-tain viral stability. The remaining parts of pVI may have aportion of pIII folded around them, sequestering them awayfrom solvent and antibody access. Contrast-enhanced electronmicrographs of negatively stained microphage particles(E. Bullitt and L. Makowski, unpublished results) exhibit aslightly enlarged diameter near the proximal end, consistentwith this model. The C-terminus of pVI has been successfullyused as a vehicle for display (50), suggesting that it is acces-sible on the virion surface.


The gene III protein is the largest and most structurallycomplex component protein of filamentous phage. It consistsof three distinct domains, separated by two glycine-rich lin-kers (residues 68–87 and 218–253), which appear to make por-tions of pIII somewhat flexible. In electron micrographs ofnegatively stained phage, the two amino-terminal domains[N1 (or D1) and N2 (or D2)] can be seen as knobby structuresat the end of the virion (see Fig. 1), and can be removed, alongwith infectivity, with subtilisin digestion (51). The carboxy-terminal 150 residues of pIII, domain D3 or CT, are proposedto interact with pVIII to form the proximal end of the viral par-ticle, as they remain associated after disruption of the viruswith detergents (38,52). The crystal structures of both N1and N2 from M13 and fd have been determined (13,53,54)(see Fig. 5), with a comparison showing N1 movement relativeto N2, with implications for infection mechanisms (see laterdiscussion).

The amino terminus of the mature pIII was the firstlocation used for the display of foreign proteins and peptideson M13 (55,56). It is still the most commonly used position.Contrary to common belief, however, polypeptides fused tothe carboxy-terminus of the M13 gene-3 minor coat proteinmay be functionally displayed on the phage surface (41). Ina phagemid display system, carboxy-terminal fusion throughoptimized linker sequences resulted in display levels compar-able to those achieved with conventional amino-terminalfusions. The details of the structure of pIII, as visualized inthe crystallographic analysis of the two N-terminal domains,suggest an innocuous structural environment remote fromthe host cell binding site and generally favorable to insertionsthat would have little impact on the function of pIII (see Sec.VII discussion).

IV.E. DNA Structure within the Viral Particle

The structure of the viral DNA within the phage particle isnot well characterized. Solid-state NMR studies of the fila-mentous phage Pf1 (57) demonstrated a uniform conforma-tion for the phosphate backbone, whereas in the Ff phage,

16 Rodi et al.

the phosphates take on a larger number of different orienta-tions. Fiber diffraction studies of the intact virion (27,32) sug-gest a lack of sufficient space within the capsid shell for aB-DNA duplex, suggesting that the DNA in the virion hasa somewhat extended conformation. Mutations resulting ina change in the net charge of the C-terminal, basic region ofpVIII (1) behaved in a manner indicating a direct but nonspe-cific electrostatic interaction between the DNA and the coatprotein. This was dramatically demonstrated by a series ofmutations in which Lys48 was converted to an unchargedamino acid. These mutants were 35% longer than wild type;the total number of basic residues in the viral interior wasunchanged (the number of proteins increased in order to com-pensate for the decreased number of charges per protein); andthe DNA structure was stretched by 35% to adapt to the low-ered charge density on the interior surface of the protein coat.

Packaging signals (or PSs) earmark viral nucleic acidsfor encapsidation and have been identified for numerousRNA- and DNA-containing viruses. The PS for filamentousphage was originally identified within the intergenic regionof f1 (shown in Fig. 2), by its ability to enable heterologousssDNA molecules to form phage-like particles (58). Bauerand Smith (59) demonstrated that the PS is located at oneend of the pV-sequestered=viral ssDNA complex, suggestingthat the PS lies exposed in this complex form, ready to initi-ate encapsulation. Hence, the DNA within filamentousphage particles is oriented within the virion such that thePS is always at the distal end (60). A hairpin structurehas been shown to exist within the virion, as it can be cross-linked with psoralen (61). The PS in its single-stranded formcan be drawn as an imperfect hairpin of 32 bp with a smallbulge. Not only can the hairpin below the bulge be deletedwith no loss of function, but so also can a portion of theupper hairpin as long as the lower is present. In addition,loop sequences at the tip of the hairpin can be altered, andshort insertions added, with no loss of function. Studies pub-lished in 1989 (12) on PS(–) mutants showed that althoughvery few phage particles were produced in the absence ofa PS, suppressor mutations allowing higher packaging


efficiency arose at high frequency. These suppressor muta-tions mapped to the coding regions of pI, pVII, and pIX,implicating their interaction with the PS during assemblyinitiation. Given that the mutant proteins function betterwith PS(þ) strains, it is assumed that the mutants havegained the ability to use another small hairpin structureas a ‘‘cryptic’’ PS. The PS of phages Ike and f1, which exhibitlittle sequence similarity, are functionally interchangeable(12). A construct with a perfect self-complementary segmentof DNA functioned poorly for packaging. Taken together,these studies point towards the conclusion that, while aduplex is essential for the function of the PS in filamentousphage, additional features are also involved.

V. FILAMENTOUS BACTERIOPHAGE LIFECYCLE

V.A. Replication of Viral DNA

Filamentous bacteriophages contain a single-stranded DNAgenome that replicates in three stages. In stage I, oncethe phage (þ) strand ssDNA (SS) is translocated into thecytoplasm subsequent to infection, bacterial host enzymessynthesize the complementary (–) strand, producing a dou-ble-stranded covalently closed supercoiled DNA product calledthe parental or replicative form (RF) DNA (SS ! RF). Stage IIoccurs when the RF DNA replicates to form a pool of approxi-mately 100 RF DNA molecules per cell (RF ! RF). In thefinal stage, the RF DNA molecules act as a template for thesynthesis of progeny single-stranded DNA phage genomes(RF ! SS).

Transcription off the (–) strand of the RF DNA occurs ina clockwise direction, as shown in Fig. 2, with gene productsproduced in the order in which they are required for phageproduction (i.e., pII, pX, pV for replication, etc.). Multipletranscripts are produced, with six mRNAs encoding pVIIIand four encoding pV, allowing for the production of largeamounts of these proteins to coat the emerging virusand intracellular ssDNA genomes, respectively (for a more

18 Rodi et al.

detailed discussion of transcription, see Ref. 62). The gene IIprotein is an endonuclease–topoisomerase that introduces aspecific nick in the (þ) strand of the RF DNA. The resulting30 end serves as a primer for synthesis of new (þ) strandsvia a ‘‘rolling circle’’ mode of replication carried out by hostcell enzymes, and terminated and circularized by pII. Theresulting (þ) strands can either be used as templates for thesynthesis of more RF DNA or be coated by pV dimers andsequestered in the cytoplasm ready to be assembled into pro-geny phage. Levels of pV thus determine the ratio of RF to (þ)strand DNA synthesis; in the presence of adequate pV, (þ)strand is coated for virus production, precluding its conver-sion to double-stranded form. Protein X is also believed tobe involved in regulation of RF=(þ) strand synthesis, as it actsas an inhibitor of pII function (5).

The pV dimer binds to DNA in a highly cooperative man-ner without marked sequence specificity at physiological pH(63–66). Protein V exhibits two distinct modes of DNA bind-ing, dictated by salt concentration, and leading to either acooperatively saturated complex through nonspecific bindingor an unsaturated complex through specific binding. Thesetwo interactions correspond to the two biological functionsof pV: sequestering (þ) strand and regulating viral mRNAtranslation (67). The nonspecific binding of pV to ssDNA pro-duces a superhelical protein–DNA complex. This complex is aflexible structure, approximately 8800 A long, and approxi-mately 80 A in diameter (64,68–70) in which the pV dimersappear to form a tightly wound left-handed helix with a pitchof about 90 A (69). The radius of gyration of the DNA in thecomplex is much lower than that of the protein, indicatingthat the DNA is closer to the axis of the structure. Each turnof the helix contains roughly eight pV dimers, with two anti-parallel ssDNA strands occupying the interior of the superhe-lix (70). A model for the pV–ssDNA complex, consistent withthese observations and with the 1.8 A resolution crystallo-graphic structure of the pV dimer, has been constructed (7).This model places the antiparallel DNA single strands withinthe grooves formed by the b-strands that loop out from thebody of the pV dimer, as illustrated in Fig. 4.


V.B. Synthesis of Viral Proteins

The replication proteins (pII, pV, and pX) are synthesized byhost cell machinery and reside within the cytoplasm. All otherviral proteins are synthesized and then inserted into eitherthe inner (pI, pIII, pVI, pVII, pVIII, pIX, and pXI) or outer(pIV) host cell membrane. Three minor capsid proteins (pVI,pVII, and pIX) are synthesized without signal peptides. Themechanisms of their membrane insertion are unknown, butall three appear to be inserted into the inner membrane withtheir carboxyl termini on the cytoplasmic side and theiramino termini on the periplasmic side. Sequence analysis pre-dicts a single membrane-spanning region in both pVII andpIX. pVI appears to have three membrane-spanning regionsand a central, positively charged region protruding into thecytoplasm, although this latter orientation is in controversy.

The gene III protein is synthesized with an 18 amino acidsignal peptide, which is removed after membrane insertion.Membrane translocation is Sec-dependent (see Ref. 71 for arecent review) and results in a single transmembrane domainlocated five residues from the carboxyl terminus, with thecarboxy-terminal residues lying within the cytoplasm (72).Most of its mass (including domains N1, N2, and most ofCT) resides in the periplasm. The membrane-anchoringdomain consists of the carboxy-terminal part of D3 and is alsoinvolved in attaching pIII to the viral particle (see Fig. 3).

Procoat (pVIII prior to cleavage by signal peptidase)appears to form dimers within the membrane by packingalong the hydrophobic face of its amphipathic helix andextending through the membrane-spanning region (73).During morphogenesis, this L-shaped conformation trans-forms into the elongated, largely a-helical structure it takeson in the intact phage particle (see Fig. 6). The major coat pro-tein pVIII was believed until recently to spontaneously insertinto the E. coli inner membrane by a Sec- and signal recogni-tion particle-independent mechanism. Recent work, however,has shown that inner membrane translocation of pVIII isdependent upon YidC, a homologue of the mitochondrialmembrane transporter Oxa 1 p (74), even though pVIII can

20 Rodi et al.

partition into the membrane in the absence of YidC (75). Onceassociated with the inner membrane, procoat appears to havea U-shaped configuration, with two transmembrane helices,extending from positions –15 to –2 and from 21 to 39 (76),bracketing an amphipathic helix that extends along the peri-plasmic surface of the inner membrane. Tyr21 and Tyr24appear to act as anchors to position the second transmem-brane helix within the membrane (77). Signal peptidase clea-vage leaves the mature pVIII in an L-shaped configuration(78) with five flexible amino-terminal residues, residues8–16 forming an amphipathic helix extending along the mem-brane surface, and residues 25–45 making up a transmem-brane a-helix that extends into the cytoplasm and include aportion of the basic carboxy-terminal region of pVIII involvedin protein–DNA interactions (79) (see Fig. 6). The extent ofthe carboxy-terminal transmembrane helix is probably not

Figure 6 Conformational shift of the gene VIII protein duringviral morphogenesis. This figure is a schematic diagram of theincorporation of coat proteins into a growing Pf1 virion. (Reprintedfrom Ref. 100, #1991 American Association for the Advancement ofScience.)


altered by signal peptidase cleavage of the amino-terminalsignal sequence. The different extents of this helix reportedby the two studies quoted above may reflect different condi-tions under which the measurements were made.

Proteins I, IV, and XI are all required for phage assem-bly, but are not present in the intact virus particle. ProteinsI and XI are synthesized without signal peptides and areinserted into the inner membrane in a SecA-dependent man-ner (80). They each span the membrane once and are orientedwith their carboxy-terminal 75 residues lying within the peri-plasm (81). Protein I has a 250 amino acid amino-terminalcytoplasmic domain, all but 10 amino acids of which are miss-ing in pXI (see Fig. 7). Plasmid-driven production of excess pIresults in a loss of membrane potential and cessation of hostcell growth, suggesting that it can form some type of channelsthrough the cytoplasmic membrane (82–84). Excess pXI doesnot appear to have such an effect. Proteins I and XI appear tointeract in the preassembly complex; pXI being the moreabundant member. The 250 amino acid cytoplasmic domainof pI contains an ATP binding site and may be activelyinvolved in assembly and extrusion of the phage particle.

The gene IV protein is synthesized with a 21 amino acidsignal peptide and is translocated into the periplasm via theSec system (16,80). The amino-terminal half of pIV forms aprotease-resistant domain exposed in the periplasm (80,85),whereas the carboxy-terminal domain mediates outer mem-brane integration and formation of a multimer composed ofbetween 10 and 12 pIV monomers that form large cylindricalclusters with an internal diameter of 80 A and probably com-prise a gated channel for phage translocation through theouter membrane (16,17,86). This channel is sufficient toaccommodate the passage of the phage particle which hasan outer diameter of roughly 65 A. A growing family of bacter-ial proteins have been identified that share significantsequence homology with the carboxy-terminal half of pIV,the ‘‘secretins,’’ which are common to the type II bacterialsecretion systems and which similarly form multimeric chan-nels (87). Protein I has been shown to associate with pIV andthis interaction has been implicated in the formation of

22 Rodi et al.

phage-specific adhesion zones, regions in which the inner andouter membranes of E. coli are in close contact (adhesionzones are more common in phage-infected cells and have beenidentified as sites of phage morphogenesis) (11,18). Recent

Figure 7 Viral morphogenesis through the pIV multimer. Thisfigure depicts the roles of various viral- and host cell-encoded pro-teins during viral assembly. (Modified from Ref. 106.) These com-plexes are believed to be located at specific adhesion zones, wherethe inner and outer bacterial membranes are in closer contact thanin nearby areas.


work has demonstrated that phage production blocks oligo-saccharide transport via pIV gated channels, providing thefirst definitive evidence for viral exit through a large aqueouschannel (88).

V.C. Viral Morphogenesis

V.C.1. Preassembly complex and the initiationof assembly

Assembly of phage particles begins when the PS interactswith the assembly complex formed by pI, pIV, and pXI andhost cell thioredoxin at localized adhesion zones (11,18) (seeFig. 7). This complex spans the inner and outer membranesand the periplasm and provides both a platform for phage par-ticle assembly and a pathway for phage particle extrusionwithout lysis of the host cell. Protein IV forms the outer mem-brane portion of the complex, pI and pXI form the inner mem-brane portion, and the amino-terminal region of pI and hostcell thioredoxin form the cytoplasmic portion of the complex.

Overexpression of pIV has no effect on host cell viability,indicating that the channel formed by pIV is closed in theabsence of other phage proteins (39). In the presence of pIand pXI and amber mutants of pVII or pIX, assemblysites are formed, followed closely by cessation of bacterialgrowth, suggesting the loss of transmembrane potential(11). Although pIV from f1 and Ike are not functionally inter-changeable, when both pIV and pI are exchanged between thetwo phage types, some heterologous phage particles areformed, suggesting that pI and pIV interact to promoteassembly (18). This interaction is confirmed by the existenceof paired compensatory mutations in the respective periplas-mic domains of pI and pIV (18). Gly355Ala=Ser mutants ofpIV that are normally conserved in all known homologs ren-der the host cell sensitive to large antibiotics and detergentsthat are normally excluded by the outer membrane (89).

Efforts to model the phage ends provide some clues to theprocess involved in the initiation of assembly. Modeling of thedistal end of the phage suggests that the PS may be sur-

24 Rodi et al.

rounded by pVIII molecules in the intact virion, rather thanby pVII and=or pIX (24). Some pVIII molecules are associatedwith pVII in the membranes of infected cells (38), suggestingthat it may be a pVII=pVIII complex that associates with thePS during initiation of assembly, followed by docking with theassembly complex.

V.C.2. Elongation

An intact assembly apparatus consists of pIV as the outermembrane component and pI=pXI as the inner membranecomponent, an intact phage ‘‘tip’’ consisting of about fivecopies each of pVII and pIX plugging the pIV channel, thePS end of the phage genome and possibly one ring of pVIIImolecules. Following formation of the assembly complex,elongation of the phage particle can begin. Assembly requiresthe presence of pIII, pVI, pVII, pVIII, and pIX in the innermembrane, as well as the host protein thioredoxin (15).Although thioredoxin in its reduced state is a potent reduc-tant of protein disulfide bonds, it is not the redox activity ofthe protein which is required for assembly as mutations ofits two active site cysteine residues do not prevent its partici-pation in phage assembly (90). One clue to its role in phageassembly is that during infection by the lytic phage T7,thioredoxin complexes with T7 DNA polymerase, conferringprocessivity on the polymerase, allowing it to polymerizethousands of nucleotides without dropping off the DNA tem-plate (91). It is conceivable that complex formation of thiore-doxin with pI enhances processivity in the filamentousphage assembly process also by stabilizing pI binding toDNA (92).

During elongation, pV is stripped from the DNA andreplaced with rings of pVIII, and simultaneously, theDNA=pVIII is exported from the host cell. Viral elongationis an ATP-driven process dependent on the presence of anintact Walker nucleotide binding motif ([AG]-XXXX-GK[ST])near the amino-terminal end of pI (62). Protein V is removedfrom the ssDNA in order for pVIII to be added, but there is noevidence of interaction between pI and pV, as gene V proteins


from various phages are interchangeable (93). It has beensuggested that ATP hydrolysis by pI can act as a source ofenergy for pV displacement from the ssDNA during elonga-tion (92). The 10 residues of pI and pXI which face thecytoplasmic surface of the inner membrane possess an amphi-philic character similar to the 10 carboxy-terminal residues ofpVIII that interact with the DNA within the intact virion,suggesting that these residues may interact with ssDNA thathas just been stripped of pV, facilitating conformational shift-ing towards interaction with pVIII (62). Because the helicalsymmetry of the DNA in the pV complex is different from thatin the viral complex, the two structures must rotate relativeto one another during the elongation process.

Studies of pVIII structure both in the intact virion andwithin detergent micelles by X-ray diffraction, solid-stateNMR, neutron diffraction, and Raman spectroscopy haveshown that a significant conformational shift must occurwithin pVIII for the protein to be incorporated into the grow-ing phage particle (27,78,79,94–99). In addition, each pVIIImolecule must now interact with the ssDNA genome and withmany other copies of pVIII, and thus, must exchange hydro-phobic interactions that anchor it within the membrane forhydrophobic interactions that stabilize it within the phageparticle (see Fig. 6). The shift from the two perpendiculara-helical segments of the membrane-bound form to the singlelong a-helix of the viral form moves the two ends of the pro-tein approximately 16 A further apart. A 16 A axial transloca-tion is precisely what is required to position the viral particleto accept further additions of pVIII rings at its proximal end.The mobility of the residues in the hinge region of themembrane-bound form is likely to enable a smooth transfor-mation from the membrane-bound form to the phage confor-mation (79,100).

Three separate research groups have performed muta-tional studies which suggest that certain small residues (suchas Ala7, Ala10, Ala18, Leu14, Gly34, and Gly38) are not easilymutated in pVIII or can only be substituted with other smallresidues (45,79,101). A model for elongation of phage particlesbased upon both NMR and X-ray diffraction studies of both

26 Rodi et al.

position and mobility has been proposed which takes intoaccount the roles of these highly conserved residues (79).

V.C.3. Termination

Termination of assembly occurs when the entire DNA mole-cule has been packaged by pVIII molecules. It seems to be arelatively inefficient process, as double-length particles thatcontain two unit-length genomes occur frequently in awild-type phage preparation (approximately 5% of the totalphage). The addition of a second genome does not representa second initiation event, as ssDNA molecules lacking a PScan be coencapsidated with a PS(þ) genome at about the samefrequency (5%) as those that contain a PS (12). Assembly ter-mination involves the addition of pVI and pIII to the tip of theparticle. Both pIII and pVI are anchored in the cytoplasmicmembrane and cannot be coimmunoprecipitated from nonio-nic detergent extracted membranes, whereas once they arewithin intact phage particles they coimmunoprecipitatefollowing the same treatment (38,102,103), inferring a stronginteraction once within the virus. Although pVI(–) and pIII(–)phages each form noninfectious ‘‘polyphage’’ and are bothlacking in pIII molecules, the pVI(–) particles are less stable,tending to unravel at the tip (11). These observations implythat pVI is needed to both stabilize the proximal end of thevirus, and attach pIII to the proximal end of the virus parti-cle. This is confirmed by the fact that anti-pVI antibodies donot interact with intact phage (38).

Protein III molecules are anchored within the cytoplas-mic membrane via residues 379–401 at their carboxyl ter-mini, with most of their residues located within theperiplasm (1–378). The amino-terminal domain (residues1–218) mediates infection (see next section), whereas thecarboxy-terminal domain (D3 or CT; residues 253–406) isinvolved in releasing the phage from the membrane and incapping the virion. Stable but noninfectious phage can beformed with a truncated pIII (residues 197–406), which com-prises a portion of the N2 domain, the second glycine-richspacer and the whole CT domain (104,105). Recent studies


with a series of carboxy-terminal fragments of increasing sizehas led to the delineation of portions of pIII that are capable ofmediating pVI incorporation into the assembling phage,release of sarkosyl-labile viral particles into the culturesupernatant, and release of sarkosyl-stable virions, respec-tively, thus dividing the D3 or CT domain into a C2 subdo-main sufficient for host cell release and a C1 subdomainrequired for virion stability (52). Coimmunoprecipitation ofa certain amount of pVIII from the membranes of infectedcells by pVI antiserum suggests that pVI is added to the par-ticle as a pVI=pVIII complex (38). It has been suggested thatonce the pVI=pVIII complex is added to the tip of the elon-gated phage, a hydrophobic patch may be exposed on thepVI surface which exhibits high affinity for the stretch of resi-dues which anchor pIII within the membrane (106). Thus,pVI=pIII interactions could replace lipid=pIII interactions,thereby releasing pIII from the inner membrane.

Alternatively, it has been noted that termination ofphage assembly results in a change in orientation of pIII rela-tive to pVIII. Prior to termination, pIII is anchored within thecytoplasmic membrane by a carboxy-terminal membraneanchor, placing the vast majority of the protein within theperiplasm (72). Protein VIII is similarly oriented with its car-boxyl terminus in the cytoplasm and amino terminus withinthe periplasm (107,108). Both prior to and after incorporationinto the virion, the two proteins interact, but after incorpora-tion the amino terminus of pIII points away from pVIII. Thischange in relative orientation between pIII and pVIII hasbeen used by Rakonjac et al. (52) to postulate a mechanismin which pIII plays the predominant role for termination ofphage assembly. Given that the segment of pIII most likelyto interact with pVIII is the membrane anchor, the rest ofthe pIII molecule is free during assembly to bend or flip inthe opposite direction from pVIII during incorporation intothe growing particle (see Fig. 8). This flipping motion on thepart of the C2 subdomain could then disrupt phage–membrane hydrophobic interactions, by allowing C2 to coverup a hydrophobic portion of the pIII carboxyl terminus, andthus resulting in release of the assembled phage from the host

28 Rodi et al.

cell. This model is corroborated by the fact that a very shortcarboxy-terminal fragment of pIII which lacks the C2 subdo-main just amino-terminal to the membrane anchor can beincorporated into the membrane-associated viral particle,but cannot detach the virus from host cells (52). Additionally,carboxy-terminal fusions of greater than 7 amino acid resi-dues have been found to prevent termination of phage assem-bly, suggesting that although the carboxyl terminus may notbe tightly packed, it may be within an enclosed environment(J. Rakonjac and P. Model, unpublished results).

V.D. The Infection Process

Filamentous phage infection is a two-step process: (i)recognition—during which the virus binds to its primary bac-terial cell surface receptor, the distal tip of the F-pilus; fol-lowed by (ii) translocation—which involves pilus retraction,capsid protein integration into the bacterial cell membrane,

Figure 8 Model for termination of viral assembly. This diagramdepicts a model proposed by Rakonjac et al. for termination ofphage assembly. A conformational shift within pIII is theorizedas the primary driving force for phage release, based upon thephenotype of a series of pIII deletion mutants. (Modified fromRef. 52.)


uncoating of viral DNA, and its concomitant translocationinto the host cell cytoplasm.

Only one or a few F-pili are present on the surface of abacterial cell. Both pilus structural proteins and the proteinsrequired for pili assembly are plasmid encoded, with the geneslying within the tra operon, or transfer region, of the conjuga-tive plasmid. Different filamentous phages have specificitiesfor different pilus types. The Ff (f1, fd, and M13) bind F-typepili, whereas Ike and I2–2 infect E. coli strains that produceN- or P-type pili. RNA phages such as R17 and Qb use the sideof the F-pilus for attachment to the host cell (109). The F-pilusconsists of a helical array of pilin subunits of 8 nm diameterwith a 2nm lumen (110). The pilin subunit is about 65–70%a-helical as determined by circular dichroism (111). Assemblyof the intact F-pilus requires the activity of 11 tra gene pro-ducts (110). This assembly process is temperature-sensitive;thus, F-specific phage fails to infect or form plaques onbacteria grown at temperatures below around 32�C.

Infection of a male E. coli cell is initiated by binding ofthe N2 domain of pIII to the tip of the F-pilus (112). This stepis followed by pilus retraction into the host cell, with pilinsubunits believed to depolymerize into the host cell innermembrane (113,114). It is not known whether phage bindinginitiates pilus retraction, or if phages are brought to the mem-brane as a consequence of normal cycles of pilus retractionand repolymerization (115). A single L70S substitution atthe last amino acid of F-pilin knocks out both pilus assemblyand Ff infection (116). Three mutations (V48A, F50A, andF60V) have been identified that do not affect pilus assembly,but reduce sensitivity to M13KO7 (an M13 derivative com-monly used as helper virus for rescue of recombinant pIIIphagemids) down to 1–3% of wild-type levels, suggesting adefect in pilus function, possibly in retraction (110).

The coreceptor for infection by Ff phage, the TolQRAcomplex, was first identified by the isolation of tolA and tolBmutants that are insensitive or ‘‘tolerant’’ to the lethal actionof a class of proteins termed ‘‘colicins,’’ which have antibioticactivity and are produced by some E. coli strains to kill non-colicin producing bacteria (117). The tolQ and tolR genes were

30 Rodi et al.

subsequently identified and shown to be not only essential tosusceptibility to certain classes of colicins, but for filamentousphage infection as well (118–120). The Tol system appears tobe involved in other types of macromolecular import, given itsstructural and sequence homologies to the TonB–ExbB–ExbDcomplex (121,122). Bacterial cells possessing functional pIIIeither from a filamentous phage infection (123) or encodedby a plasmid (10) show increased tolerance to the effect ofE-type colicins, suggesting that pIII and E colicins have iden-tical or closely adjacent binding sites on the Tol complex. Inaddition, overexpression of pIII, its amino-terminal fragment(10,124) or proteins possessing the TolA D3 domain (125,126)induces outer membrane leakiness in E. coli, leading to loss ofperiplasmic proteins.

TolA, TolQ, and TolR form a complex within the innermembrane of the bacterial host cell as shown schematicallyin Fig. 9. The N-terminus of TolA anchors the protein to the

Figure 9 The Ff bacteriophage infection process. This figureshows a schematic view of phage infection demonstrating howpilus-mediated separation of N1 from N2 at the amino terminusof pIII frees up N1 for interaction with the D3 domain of thecoreceptor TolA, thus mediating viral entry into the host cell.(Reprinted from Ref. 14, #1999 Elsevier Science.)


inner membrane, whereas the C-terminus, D3, is associatedwith the outer membrane, connected by an extended helicalregion, D2 (127). Analysis of the crystal structures of thetightly interacting N1 and N2 domains of pIII (54) (Fig. 5)and the complex between N1 of pIII and the carboxy-terminaldomain D3 of its coreceptor TolA (14) (see Fig. 10) demon-strates that during the infection process, the interaction

Figure 10 Structure of the N1 domain of pIII cocrystallized withthe D3 domain of TolA. The cocrystal structure of the D3 domain ofthe coreceptor TolA with the amino-terminal domain of pIII hasbeen solved to 1.85 A resolution (14) (PDB accession code 1TOL).Note that the amino-terminal residue of pIII (arrow) lies far awayfrom the interaction surface between the two proteins.

32 Rodi et al.

between the N1 and N2 domains of pIII is replaced by theinteraction of the D3 domain of TolA with N1. Despite a lackof topological similarity between pIII–N2 and TolA–D3 (com-pare Figs. 5 and 10), both domains interact with the sameregion of the pIII–N1 domain and bury comparable accessiblesurface areas (1768 A2 for pIII–N1=TolA–D3 vs. 2154 A2 forpIII–N1=pIII–N2) (14). Fourteen of the 21 residues of thepIII–N1 domain that are involved in the interactions withTolA–D3 are also involved in interactions with pIII–N2.Infection of F� bacteria with wild-type phage can be achievedat low levels (4–5 orders of magnitude lower than the rate forFþ hosts) by treatment of the bacteria with 50mM Ca2þ, pre-sumably by altering the outer membrane enough to mediateexposure of TolA–D3 for interaction with the N1 domainof pIII.

The mechanism by which capsid proteins are integratedinto the host cell cytoplasmic membrane and viral DNA isuncoated and translocated into the bacterial cytoplasm is lar-gely unknown. Gene VIII proteins which have become asso-ciated with the inner membrane can later be reutilized inthe assembly of progeny phage particles, as their insertioninto the membrane occurs in a manner which gives themthe same topology as newly synthesized pVIII molecules(128–130). Penetration of the DNA has been demonstratedto require host cell TolQRA proteins (126). It has been sug-gested that the CT domain of pIII may be involved in the for-mation of an entrance pore for DNA translocation (131),perhaps as a mirror image of the mechanism by which pIIImediates termination of assembly (see Fig. 8) (52), with a con-formational change of the CT domain of pIII uncapping thevirion, exposing hydrophobic surfaces of pIII, pVI, and pVIII,followed by membrane integration. This would be analogousto the mechanism of entry of eukaryotic viruses into hostcells, via the unmasking of hydrophobic fusogenic peptides(132). In support of this hypothesis, it has been shown thatthe insertion of a b-lactamase domain between the aminoand carboxyl termini of pIII (thus disrupting their properdistance) decreases infectivity by two orders of magnitude(133).


VI. PHAGE LIBRARY DIVERSITY

VI.A. Efficiency as a Biological Strategyfor Survival

Considering the effect of insertion mutations on M13 requiresan examination of the relationship between M13 and its E. colihost. M13 is not a lytic phage. Rather, it parasitizes the host,being carried from generation to generation and producinganywhere from 200 to 2000 progeny phage per cell per doublingtime (2,3). As discussed above, this phage production repre-sents a serious metabolic load for the infected E. coli, corre-sponding to 1–5% of total protein synthesis and resulting in areduction in cell growth to perhaps 30–50% of uninfected cells.Given this negative effect on host growth, any host mutationthat resulted in resistance to phage would seem to be highlyfavorable, and resistant host should quickly outgrow infectedhost. These mutations are not observed to occur, suggestingthat the phage has evolved strategies for preventing its hostfrom developing resistance. What form do these strategiestake? Mutations that knock out expression of any of the viralgenes (with the exception of pII) result in killing of the host cell(2,134). This effect appears to be due to the accumulation ofphage-encoded proteins III and I in the cytoplasmic membranethat results in the degradation of host cell membranes. Theseobservations suggest that any mutation in either the host cellgenome or phage-encoded proteins that results in the haltingor even the slowing of phage morphogenesis may lead to thebuildup of pI and pIII in the host cell, and subsequently tothe death of the host. Similarly, anymutation in a viral proteinthat blocks or slows phage assembly in such a way as to allowthe accumulation of pI or pIII may also lead to host cell death.If this is the case, inserts that slow phage production may berapidly censored from a phage-displayed library if they resultin the buildup of pI and pIII in the host cell membrane. M13appears to have evolved to prevent the development of resis-tance in its natural host: any mutation that significantly slowsits production represents a fatal mutation. A detailed examina-tion of the diversity and censorship patterns of phage displaylibraries (135,136) reinforces this hypothesis.

34 Rodi et al.

VI.B. Phage Population Diversity

A number of groups have investigated the biochemical diver-sity of phage display constructs using various methods includ-ing restriction digestion pattern analysis of small numbers ofgroup members and colony hybridization with primers(137–142). Statistical methods have been developed and usedto quantitate and annotate the sequence diversity of combina-torial peptide libraries on the basis of small numbers (100 or200) of sequences. Application of these methods in the analysisof commercially available M13 pIII-based phage displaylibraries (136) demonstrated that these libraries behave statis-tically as though they correspond to populations containingroughly 4.0% of the random dodecapeptides and 7.9% of therandom constrained heptapeptides that are theoretically pos-sible within the phage populations. Analysis of amino acidoccurrence patterns in these libraries shows no demonstrableinfluence on sequence censorship by E. coli tRNA isoacceptorprofiles or either overall codon or Class II codon usage pat-terns, suggesting no metabolic constraints on recombinantpIII synthesis. This is in contrast to a clear effect of metaboliclimitations to the diversity of pVIII libraries (135). The pIIIlibraries exhibit an overall depression in the occurrence ofcysteine, arginine, and glycine residues and an overabundanceof proline, threonine, and histidine residues, and position-dependent amino acid sequence bias that is clustered at threepositions within the inserted peptides of the dodecapeptidelibrary, þ1, þ3, and þ12 downstream from the signal pepti-dase cleavage site. These sequence limitations can primarilybe attributed to two steps during viral assembly: signal pepti-dase cleavage and incorporation of the recombinant proteininto the growing virion from the bacterial inner membrane.

VII. BIOLOGICAL BOTTLENECKS: SOURCESOF LIBRARY CENSORSHIP

VII.A. Protein Synthesis

Inserts into the minor structural proteins of M13 are not likelyto significantly disrupt or slow their synthesis, and this has


been confirmed for combinatorial peptide libraries displayed onpIII (136). However, the sheer numbers of pVIII molecules thatmust be synthesized for viral production put the major struc-tural protein of M13 in a separate category. The protein synth-esis apparatus of the infected host cell produces from 5� 105 to5� 106 copies of pVIII per doubling time. It is well establishedthat rare codons are used by the E. coli bacterium to regulatethe rate of synthesis of regulatory proteins (143) and thatmRNAs dependent on rare codons can inhibit protein synthesisby robbing the cell of minor tRNA isoacceptors (144,145). Wild-type pVIII does not include any of the 10 rarest codons inE. coli, presumably because their presence would greatlyretard the production of pVIII and limit virion production.

Rodi and Makowski (135) analyzed the relative occurrenceof codons in a pVIII pentapeptide library constructed with a 32codon code. The insert was of the form (NNK)5, where N referstoanequimolarmixtureofall fourbasesandKanequimolarmix-ture of G and T. Seven amino acids have multiple codons in alibrary of this form. For four of them, a statistically significantcorrelation between frequency of codon use in the library andabundance of their respective tRNAwas observed. These resultsindicate that the presence of rare codons in an insert into pVIIIsignificantly decrease the likelihood that the insert will be suc-cessfully expressed on the surface of the phage. Proteins VIIIandVare the only phageproteins synthesized in very largenum-bers as required for their functions. All other phage proteins areexpressed at relatively low levels. These facts are reflected in thelack of rare codons in these proteins. Neither pVIII nor pV haveany copies of the rarestE. coli codonswhereas there aremultiplecopies of these rare codons in pI and pIII. The presence of theserare codons undoubtedly contributes to the low expression levelsof these proteins in host cells. Consequently, the presence of rarecodons inan insertedsequence is likely tohaveasignificanteffecton the levels at which a protein is expressed.

VII.B. Protein Insertion in the Inner Membrane

It has been demonstrated multiple times that excess positivecharges at the amino-terminal end of a membrane protein

36 Rodi et al.

reduces its transport across the membrane (146,147), due tothe electrical component of the proton motive force (pmf)(148). This predicts censorship of pIII or pVIII libraries, limit-ing the positive charges included in inserts at the amino ter-minus. In addition, it has been demonstrated that anarginine-specific restriction exists that may be due to theinteraction of the translocating protein with the SecY protein.Rodi et al. (136) observed a net decrease in charge in two com-binatorial peptide libraries displayed at the amino terminusof pIII and noted that this effect was due solely to a decreasein arginine residues, as no underabundance of lysine residueswas observed. Peters et al. (149) constructed multiple polyar-ginine mutants of pIII and found a variable reduction inexport of phage particles into the media. No polylysinemutants were constructed in this study. Furthermore, theirattempts to rescue export by the addition of negativelycharged residues (i.e., glutamic and aspartic acids) were tono avail. Export rescue of the arginine-rich mutants wasachieved, however, by infection of prlA mutants. The prlAphenotype has been shown to be the result of mutationswithin the SecY protein, the largest subunit of the SecYEGtranslocase complex. Specifically, prlA mutants possess aloosened association among the subunits, which facilitatesATP-dependent coinsertion of a portion of SecA (the ATPasesubunit loosely associated with SecYEG) with the preprotein(150). Furthermore, prlA4 relieved translocation blockagecaused by certain folded structures. It has been demonstratedthat in order to be translocated, secretory proteins need to beat least partially unfolded (151,152). These observations sug-gest that mutants that rescue export of the arginine-richmutants also provide for export of bulkier, folded structuresnot translocated by wild-type strains.

VII.C. Protein Processing

After insertion into the inner host cell membrane, the signalpeptides of pIII and pVIII must be cleaved by signal peptidasein order for the proteins to be available for incorporation intothe assembling virus particle. There is evidence for sequence


preferences in the first two or three positions downstreamfrom the cleavage site (153), a favored position for incorpora-tion of foreign peptides or proteins. Rodi et al.(136) observed astatistically unexpected peak of proline residues at the þ 3andþ 12 positions in a pIII dodecapeptide library and theþ 3 and þ 4 positions in a constrained heptapeptide library.In addition, it has been reported that proline at the þ 1 posi-tion acts as an inhibitor of the signal peptidase enzyme whichcleaves the leader sequence from the mature protein subse-quent to membrane insertion (146,154–156). The mild censor-ship of proline at þ 1 observed in the dodecapeptide pool maybe explained on the basis of this inhibition.

Furthermore, except for the first position, there is a sig-nificant overabundance of proline in combinatorial peptidesdisplayed at the amino terminus of the mature pIII (136).There is also a dramatic overabundance of proline over mostof the length of the peptides in the pIII-displayed libraries(136). This correlates with both the weak preference for pep-tides with a high propensity for b-turns in these populationsand a preference of signal peptidase for b-turn conformations.Whether the observed overabundance of proline in theselibraries is due to the preferred three-dimensional motif forsignal peptidase substrates or to a later step in morphogen-esis cannot be determined from this data alone. Maliket al. (35) found significantly reduced processing of a peptideinsert which had a high propensity for a-helix formation. Allthese data point to the conformational nature of the incorpo-rated peptide being an influence on the efficiency of signalcleavage and consequent inclusion within the display librarypopulation.

VII.D. Display in the Periplasm

E. coli has a highly effective dsb system for formation ofdisulfide bonds in the periplasm. Consequently, inserts atthe amino terminus of any of the structural proteins may becrosslinked prior to assembly if they contain a cysteine resi-due. Work done by Haigh and Webster (73) has shown that,prior to incorporation into the growing virion, the close

38 Rodi et al.

proximity of pVIII molecules within the inner membraneresults in a high degree of crosslinking between singlecysteines in different pVIII molecules, thus precluding theirparticipation in viral morphogenesis. A similar effect hasbeen observed in pIII libraries. The almost complete absenceof odd numbers of cysteine residues in an amino-terminal pIIIdisplay library (157) resulted in significant censorship of thatlibrary. This crosslinking may be intramolecular, involvingone of the other cysteines in the pIII (and potentially result-ing in the misfolding of the pIII), or it may be intermolecular,resulting in dimerization which would preclude incorporationof the pIII into the growing virus particle.

VII.E. Viral Morphogenesis

Inserts that interfere with the protein–protein interactionsthat occur during viral morphogenesis have the potential todecrease or eliminate viral production. In spite of a great dealof evidence suggesting that inserts can disrupt viral assem-bly, the mechanism of this disruption is not well characterizedfor any specific case. This may involve the interaction of viralproteins as they move actively or passively through thepI–pXI–thioredoxin complex associated with the inner mem-brane, or through the pIV pore in the outer membrane.Peptides or proteins displayed at the amino terminus of pIIIor pVIII are exposed to the periplasm prior to assembly. Fromthat position, it would seem relatively unlikely for them todisrupt protein–protein interactions involving pI or pXI. Theycould, however, pose a problem for phage during extrusionthrough the outer membrane pIV pore. Display on pIII maynot be limited by the pIV pore, since pIII appears to be‘‘dragged’’ through the outer membrane at the tail end ofthe virus particle. Display on pVIII, however, could beseverely limited by the pore.

Early work indicated that the display of peptides onpVIII (when present on every copy of pVIII) was limited toaround 6 amino acids (158,159). Iannolo et al. (34) reportedthat the majority of six residue inserts, 40% of 8 residueinserts, 20% of 10 residue inserts, and only 1% of 16 residue


inserts, were tolerated within pVIII. This seems consistentwith the demonstration that a peptide inserted near theamino terminus of mature pVIII occupies a shallow grooveon the surface of the phage particle and that this groove iscapable of accommodating about eight residues (32). The pre-sumptive conclusion is that peptides that cannot be accommo-dated by the shallow groove will not allow passage of theparticle through the outer membrane pore.

Other data, however, suggest that this may not be thecorrect interpretation. In hybrid display systems, it is possibleto display large proteins at the amino terminus of pVIII, aslong as both mutant and native pVIII are present. If theextrusion of the virus through the outer membrane is sorestrictive as to prevent the display of relatively short pep-tides on every pVIII, how can the extrusion of phage display-ing a limited number of large proteins on pVIII be possible?Given how little we know about the process, we cannotexclude the possibility that the mechanics of viral extrusionwill allow for export of a few large displayed proteins butexclude export of virions displaying many short peptides,somewhat akin to moving large irregular pieces of furniturethrough a small doorway.

Equally compatible with these data is the theory that theouter membrane pore is not so tight as to exclude virions dis-playing peptides about 10 or more amino acids in length, butthat the exclusion of longer peptides may be due to limitationsimposed by other phases of the phage life cycle. It is possiblethat the metabolic demand upon E. coli becomes limiting forlong inserts if present within every copy of pVIII (see discus-sion above). It has also been pointed out that pVIII is involvedin special interactions at the two ends of the virus particle(24,160)—those needed to attach pVI, pVII, and pIX to theparticle—and that these interactions may have specialrequirements on either length or nature of the insert that can-not be readily met. Finally, the interaction of pVIII with pIduring assembly (18) may not involve every pVIII. An insertthat disrupts this interaction may represent a fatal mutationif every pVIII harbors it, but not in a hybrid phage with nativepVIII present to perform that function.

40 Rodi et al.

VII.F. Infection Process

For those viral particles that escape host cell one (i.e., the cellthat acquired the electroporated DNA) via successful virionassembly, the last step of growth is reinfection of Fþ bacterialcells. This process is initiated by the attachment of the N2domain of pIII to the tip of the F pilus. Following pilus retrac-tion, the N1 domain of pIII (to which the random peptide isattached) interacts with the third domain of the bacterialsurface receptor TolA (TolA–D3). The cocrystallization of thesetwo purified domains, as shown in Fig. 5, delineated the inter-action surfaces required for infection by M13 (53). Given theposition of the amino terminus of pIII, only a fully extendeddodecapeptide would be likely to impart significant interfer-ence of the infection process. The position of both the TolAbinding site and the intermolecular interface with pIII–N2lie opposite the amino terminus, explaining why fusions ofpeptides and proteins to the amino terminus of pIII do not pre-clude phage infection (53). A possible rationale for the largeoverabundance of proline at þ12 is that a proline residue atthat position (arrow in Fig. 10) might direct the insertedpeptide away from the binding site between TolA–D3 andN1 of pIII, minimizing loss of infectivity.

VIII. QUANTITATIVE DIVERSITY ESTIMATION

The biology of the phage–host system as reviewed above actsas a censor that inevitably limits the diversity of a phage-displayed library. Since the utility of a library is proportionalto the diversity of the library, it is important to have quanti-tative measures of diversity to assess library quality.

As a surrogate for a true measure of diversity, thenumber of independent clones in a library and the inferrednumber of copies of each peptide=protein are sometimesquoted as a measure of library complexity (e.g., Ref. 142).Scott and Smith (56) calculated the probability of peptidesbeing present in a library of 2.3� 106 clones assuming Poissonstatistics and equal probability of occurrence for all possibleclones. Cwirla et al. (137) recognized that the apparent


diversity of a peptide library will be limited by the fact thateach of the 20 amino acids is coded for by different numbersof distinct codons. They further observed that roughly mostamino acids occurred at most positions in their hexapeptidelibrary, leading them to conclude that viral morphogenesisdid not impose severe constraints on the diversity of theirlibrary. The effect of viral morphogenesis is, in fact, observa-ble and significant but not severe (136). DeGraaf et al. (138)carried out a more extensive analysis of diversity in aphage-displayed library of random decapeptides. They ana-lyzed the sequence of 52 clones selected at random from apopulation of 2� 106 individual clones and demonstrated thatthe frequency of amino acid occurrence in this library had arough correlation with that expected from the number of dis-tinct codons corresponding to each amino acid. They furtheridentified the presence of 250 of the 400 dipeptides theoreti-cally possible in the 52 decapeptides selected and analyzed.Since only 468 dipeptides are included in this limited popula-tion of 52 decapeptides, this observation is not significantlydifferent from that expected from random sampling. Althougheach of these analyses was motivated by the need to measurethe diversity, or complexity, of a peptide library, each fallsshort of a true quantitative measure of library diversity.

There are two possible approaches one can take to definethe diversity of a population with multiple copies of individualmembers—‘‘technical diversity,’’ or completeness, correspondingto the percentage of possible members of a population that existat any copy number within a population; and ‘‘functional diver-sity,’’ which additionally takes into account the copy numbersof each distinct member of the population. In the latter scenario,if the copy numbers of the members present in the populationare dramatically different, the diversity is intrinsically lower.Real phage-displayed peptide populations invariably containunequal numbers of different peptides and any useful measureof sequence diversity must take this into account. Experimentsthat utilize sequence information from limited numbers of popu-lation members to estimate peptide population diversity cannotprovide accurate estimates of completeness, since very raremembers of the population will inevitably go unsampled.

42 Rodi et al.

Limited sequence information, however, is capable ofestimating the functional diversity of a peptide library.Makowski and Soares (161) introduced an analytical expres-sion for the ‘‘functional’’ diversity of a population of peptides,not only demonstrating that this expression is consistent withintuitive expectations for the properties of population diver-sity, but providing a means to calculate the diversity on thebasis of a limited number of peptide sequences. Whereas,the diversity of a population containing equal numbers of halfof all possible peptides (and no copies of the remaining half) isintuitively equal to 50% (or 0.5), the diversity of populationscontaining unequal numbers of different peptides is less easyto calculate. Their measure of library diversity is directlylinked to the probability of selecting the same library membertwice during random selection from the population. As aresult, the ‘‘functional’’ diversity is equal for all populationsin which this probability is the same.

For a population in which there is a theoretical maxi-mum of N possible members, the diversity, d, is defined as:

d ¼ 1=ðNX

kP2kÞ

where the sum is over all possiblemembers, k, andpk is theprobability of the k-th member being selected in any randomselection from the population—a directmeasure of the relativeabundance of the peptides. Note that for any population,P

k Pk ¼ 1. If all possiblemembersarepresent inequalnumbers(equal probability of being chosen), then pk ¼ 1=N for all k. Thesum in the equation is then equal to N(1=N2) ¼ (1=N), andd ¼ 1.0 as expected. If half (N=2) of the members are presentin equal numbers and the other half are missing, then for halfthe population, pk ¼ 2=N, and for the other half, pk ¼ 0. Thesum in Eq. (1) is then (N=2) (2=N)2 ¼ 2=N, and it follows that d¼ 0.5 as one would intuitively expect.

For combinatorial peptide libraries, this diversity can bereadily estimated from the frequency of occurrence of eachamino acid at each position in the library (161). Furthermore,by combining a quantitative estimation of the diversity ofpeptide libraries with peptide sequence pattern analysis, it


is possible to measure the impact of various steps in the virallife cycle on library quality (136). According to the diversitydefinition of Makowski and Soares (161), the sequence diver-sity of a population of peptides in which each amino acid resi-due is present at each position one-twentieth or 5% of the timeis estimated to be 1.0 (100%). However, the relative levels ofeach of the 20 amino acids at each position within the randompeptides is not random, but is determined by the numbers ofcorresponding codons within a 32 (or 61) codon system notjust a simple factor of 1=20. As some of the amino acids arecoded by two or three codons, this gives an enhanced abun-dance of certain residues such as leucine (six codons) andglycine (four codons). Let us use a genetically random dodeca-peptide insert at the amino terminus of pIII as a test case fordiversity quantitation and censorship source assignment. Anin silico computationally constructed dodecapeptide librarybased upon a 32 codon code behaves statistically as thoughonly 11.8% of all possible peptides are present. These valuesindicate that due to the redundant nature of the genetic code,the diversity of a random dodecapeptide library drops from100% to 11.8%, a decrease in diversity greater than thatcaused by other host-virus factors (136).

The second largest quantitative effect on diversity ofcombinatorial peptide libraries on M13 is the almost com-plete absence of odd numbers of cysteine residues (141,157).McConnell et al. (141) and Lowman and Wells (162) postu-lated that unpaired cysteine residues within an amino-terminal extension on pIII could form disulfide bonds withone of the four native cysteine residues in the N1 domain ofpIII and partially or completely interfere with the infectionprocess. Within 100 randomly selected peptide sequencesfrom a dodecamer library, although one would expect to see37.5 cysteine residues as dictated by codon frequency, only11 residues were observed (136). Work done by Haigh andWebster (73) has shown that pVIII molecules are close enoughtogether within the inner membrane prior to incorporationinto the growing virion that single cysteine residues have atendency to crosslink between different pVIII molecules, pre-cluding their participation in viral morphogenesis. A similar

44 Rodi et al.

scenario may be possible with unpaired cysteine residueswithin pIII molecules in a pentavalent fusion display library,where intermolecular disulfide bridges could interfere witheither proper morphogenesis and=or infection. It is also possi-ble that single cysteine residues may occasionally be dis-played in a structural context that allows them to avoid thecrosslinking activity of the E. coli periplasmic dsb system.Although cysteine is one of only 20 amino acids, 63% of allpossible 12-mer peptides would be expected to have an oddnumber of cysteine residues. Censorship of all odd-numberedcysteine-containing peptides would reduce by 63% the 11.8%diversity remaining after accounting for codon bias, droppingthe 11.8% further down to 4.4% of the total number possiblein a random dodecapeptide pIII library.

Finally, an antiarginine bias within the first half of therandom peptide sequence contributes a relatively minor addi-tional bias (see discussion above). If we approximate that biasto maximal levels (i.e., no arginine residues at þ1), we loseanother 0.2% in diversity and are down to 4.3% total. Anassumption of 100% loss of glycine at þ2 removes another0.2%, arriving at a value of 4.2% diversity, well within theestimated 4.0 � 1.6% number calculated in Rodi et al. (136).

The obligate steps of membrane insertion and signalpeptidase cleavage result in patterns of censorship that arereflected in the statistical properties of the libraries. Quantita-tively, however, it is viral morphogenesis that is the predomi-nant biological source of sequence bias within the randomdodecapeptide library analyzed above. Due to the presence ofodd numbers of cysteine residues in the dodecapeptide library,as many as 63% of pIII molecules are either trapped at theinner membrane unable to participate in viral assembly dueto crosslinking via disulfide bonds or are unable to undergosuccessful infection due to improperly folded pIII–N1 domains.

IX. IMPROVED LIBRARY CONSTRUCTION

The quantitative analysis described above indicated that, forcombinatorial peptide libraries displayed on pIII, the nature


of the genetic code has the greatest impact on library diver-sity; followed by the censorship of odd numbers of cysteinesand an amino-terminal censorship of arginines. Comparisonof the amino acid composition of in silico constructedhuman proteome-derived and codon-dictated libraries along-side actual in vivo-assembled random peptide librariesdemonstrates that the sequence biases seen in real-life phagelibraries do not bias them either towards or away from thecomposition or statistical properties of real proteins as com-pared to computationally constructed random libraries(136). These biases simply increase the chance of some motifsand decrease the chance of other motifs being observed. Thepredominant reasons for this are the methods of construction(i.e., representation is strictly codon based) and the biologicalbottlenecks imposed primarily by the viral morphogenesisprocess which necessitates translocation and assembly ofthe virus through two membranes and the periplasmic spaceof E. coli.

A more diverse population could be obtained by usingspecifically constructed trinucleotide cassettes for the synth-esis of random inserts. By incorporating the trinucleotidescorresponding to the most common tRNA isoacceptors foreach amino acid in the ratio of amino acid occurrences asmeasured within the human genome, even given the assem-bly-associated sequence censorship, a library of random dode-capeptides could be constructed with a functional diversityapproaching 33%, an improvement of threefold over presentcapabilities. Insertion of a sequence that is a good substratefor signal peptidase between the signal peptidase cleavagesite and the peptide library may further increase the func-tional diversity of the library towards a value approaching37%. Propagation within a prlA host with its loosened trans-locase complex would also contribute to a more complex popu-lation of displayed peptides on M13, although a quantitationof this effect is difficult to estimate.

Consideration of the phage–host biology provides signifi-cant guidance for the design of improved phage-displayedlibraries. Given the impact and widespread application ofthe existing libraries, incorporation of improvements based

46 Rodi et al.

on an understanding of the biology of the phage–host systemshould substantially improve the success rates for experi-ments involving the use of these libraries as outlined in theremainder of this volume.

REFERENCES

1. Symmons MF, Welsh LC, Nave C, Marvin DA, Perham RN.Matching electrostatic charge between DNA and coat proteinin filamentous bacteriophage. Fibre diffraction of charge-deletion mutants. J Mol Biol 1995; 245:86–91.

2. Marvin DA, Hohn BA. Filamentous bacterial viruses. Bacter-iol Rev 1969; 33:172–209.

3. Model P, Russel M. Filamentous bacteriophage. In: CalendarR, ed. The Bacteriophages. Vol.2. NY: Plenum Publishing,1998:375–456.

4. Haigh NG, Webster RE. The pI and pXI assembly proteinsserve separate and essential roles in filamentous phageassembly. J Mol Biol 1999; 293:1017–1027.

5. Fulford W, Model P. Gene X of bacteriophage f1 is requiredfor phage DNA synthesis-mutagenesis of in-frame overlap-ping genes. J Mol Biol 1984; 178:137–153.

6. Terwilliger TC. GeneV protein dimerization and cooperativityof binding of poly (dA). Biochemistry 1996; 35:16652–16664.

7. Skinner MM, Zhang H, Leschnitzer DH, Guan Y, Bellamy H,Sweet RM, Gray CW, Konings RNH, Wang AH-J, TerwilligerTC. Structure of the gene V protein of bacteriophage f1 deter-mined by multiwavelength X-ray diffraction on the seleno-methionyl protein. ProcNatl Acad Sci USA 1994; 91:2071–2075.

8. Prompers JJ, Folmer RHA, Nilges M, Folkers PJM, KoningsRNH, Hilbers CW. Refined solution structure of theTyr41!His mutant of the M13 gene V protein: A comparisonwith the crystal structure. Eur J Biochem 1995; 232:506–514.

9. Michel B, Zinder ND. Translational repression in bacterio-phage f1: characterization of the gene V protein target onthe gene II mRNA. Proc Natl Acad Sci USA 1989; 86:4002–4006.


10. Boeke JD, Model P, Zinder ND. Effects of bacteriophage f1Gene III protein on the host cell membrane. Mol Gen Genet1982; 186:185–192.

11. Lopez J, Webster RE. Assembly site of bacteriophage f1corresponds to adhesion zones between the inner and outermembranes of the host cell. J Bacteriol 1985; 163:1270–1274.

12. Russel M, Model P. Genetic analysis of the filamentous bac-teriophage packaging signal and of the proteins that interactwith it. J Virol 1989; 63:3284–3295.

13. Holliger P, Riechmann L, Williams RL. Crystal structure ofthe two N-terminal domains of g3p from filamentous phagefd at 1.9 A: evidence for conformational lability. J Mol Biol1999; 288:649–657.

14. Lubkowski J, Hennecke F, Pluckthun, Wlodawer A. Filamen-tous phage infection: crystal structure of g3p in complex withits coreceptor, the C-terminal domain of TolA. Structure1999; 7:711–722.

15. Russel M. Filamentous phage assembly. Mol Microbiol 1991;5:1607–1613.

16. Russel M, Kazmierczak B. Analysis of the structure andsubcellular location of filamentous phage pIV. J Bacteriol1993; 175:3998–4007.

17. Kazmierczak BI, Mielke DM, Russel M, Model P. pIV, a fila-mentous phage protein that mediates phage export across thebacterial cell envelope, forms a multimer. J Mol Biol 1994;238:187–198.

18. Russel M. Protein–protein interactions during filamentousphage assembly. J Mol Biol 1993; 231:689–697.

19. Feng J-N, Model P, Russel M. A trans-envelope protein com-plex needed for filamentous phage assembly and export. MolMicrobiol 1999; 34:745–755.

20. MarvinDA.Model-building studies of Inovirus: genetic variations ona geometric theme. Int J Biol Macromol 1990; 12:125–138.

21. Berkowitz SA, Day LA. Mass, length, composition and struc-ture of the filamentous bacterial virus fd. J Mol Biol 1976;102:531–547.

48 Rodi et al.

22. Specthrie L, Bullitt E, Horiuchi K, Model P, Russel M,Makowski L. Construction of a microphage variant of fila-mentous bacteriophage. J Mol Biol 1992; 228:720–724.

23. Simons GF, Konings RN, Schoenmakers JG. Genes VI, VIIand IX of phage M13 code for minor capsid proteins of thevirion. Proc Natl Acad Sci USA 1981; 78:4194–4198.

24. Makowski L. Terminating a macromolecular helix: structuralmodel for the minor proteins of bacteriophage M13. J Mol Biol1992; 228:885–892.

25. Makowski L, Caspar DLD. The symmetries of filamentousphage particles. J Mol Biol 1981; 145:611–617.

26. Opella SJ, Stewart PL, Valentine KG. Protein structure bysolid-state NMR spectroscopy. Q Rev Biophys 1987; 19:7–49.

27. Glucksman MJ, Bhattacharjee S, Makowski L. Three-dimensional structure of a cloning vector: X-ray diffractionstudies of filamentous bacteriophage M13 at 7 A resolution.J Mol Biol 1992; 226:455–470.

28. Marvin DA, Hale RD, Nave C, Citterich MH. Molecular mod-els and structural comparisons of native and mutant class Ifilamentous bacteriophages Ff (fd, f1, M13), If1 and Ike. JMol Biol 1994; 235:260–286.

29. Marvin DA. Filamentous phage structure, infection andassembly. Curr Opin Struct Biol 1998; 8:150–158.

30. Greenwood J, Hunter GJ, Perham RN. Regulation of bacter-iophage length by modification of electrostatic interactionsbetween coat protein and DNA. J Mol Biol 1991; 217:223–227.

31. Terry TD, Malik P, Perham RN. Accessibility of peptides dis-played on filamentous virions: susceptibility to proteases. BiolChem 1997; 378:523–530.

32. Kishchenko G, Batliwala H, Makowski L. Structure of a for-eign peptide displayed on the surface of bacteriophage M13.J Mol Biol 1994; 241:208–213.

33. Brown JH, Jardetzky TS, Gorga JC, Stern LJ, Urban RG,Strominger JL, Wiley DC. Three-dimensional structure ofthe human class II histocompatibility antigen HLA-DR1.Nature (London) 1993; 364:33–39.


34. Iannolo G, Minenkova O, Petruzzelli R, Cesareni G. Modify-ing filamentous phage capsid: limits in the size of the majorcapsid protein. J Mol Biol 1995; 248:835–844.

35. Malik P, Terry TD, Gowda LR, Langara A, Petukhov SA,Symmons MF, Welsh LC, Marvin DA, Perham RN. Role ofcapsid structure and membrane protein processing in deter-mining the size and copy number of peptides displayed onthe major coat protein of filamentous bacteriophage. J MolBiol 1996; 260:9–21.

36. Petrenko VA, Smith GP, Gong X, Quinn T. A library oforganic landscapes on filamentous bacteriophage. ProteinEng 1996; 9:797–801.

37. Malik P, Terry TD, Bellintani F, Perham RN. Factors limit-ing display of foreign peptides on the major coat protein offilamentous bacteriophage capsids and a potential role forleader peptidase. FEBS Lett 1998; 436:263–266.

38. Endemann H, Model P. Location of filamentous phage minorcoat proteins in phage and in infected bacteria. J Mol Biol1995; 250:496–506.

39. Marciano DK, Russel M, Simon SM. An aqueous channel forfilamentous phage export. Science 1999; 284:1516–1519.

40. Fuh G, Pisabarro MT, Li Y, Quan C, Lasky LA, Sidhu SS.Analysis of PDZ domain-ligand interactions using carboxyl-terminal phage display. J Biol Chem 2000; 275:21486–21491.

41. Fuh G, Sidhu SS. Efficient phage display of polypeptidesfused to the carboxy-terminus of the M13 gene-3 minor coatprotein. FEBS Lett 2000; 480:231–234.

42. Roberts LM, Dunker AK. Structural changes accompanyingchloroform-induced contraction of the filamentous phage fd.Biochemistry 1993; 32:10479–10488.

43. Oh JS, Davies DR, Lawson JD, Arnold GE, Dunker AK. Iso-lation of chloroform-resistant mutants of filamentous phage:localization in models of phage structure. J Mol Biol 1999;287:449–457.

44. Iannolo G,MinenkovaO, Gonfloni S, Castagnoli L, Cesareni G.Construction, exploitation and evolution of a new peptide

50 Rodi et al.

library displayed at high density by fusion to the major coatprotein of filamentous phage. Biol Chem 1997; 378:517–521.

45. Weiss GA, Wells JA, Sidhu SS. Mutational analysis of themajor coat protein of M13 identifies residues that controlprotein display. Protein Sci 2000; 9:647–654.

46. Houbiers MC, Spruijt RB, Demel RA, Hemminga MA, WolfsCJAM. Spontaneous insertion of gene 9 minor coat proteinof bacteriophage M13 in model membranes. Biochim BiophysActa 2001; 1511:309–316.

47. Houbiers MC, Wolfs CJAM, Spruijt RB, Bollen YJM,Hemminga MA, Goormaghtigh E. Conformation and orienta-tion of the gene 9 minor coat protein of bacteriophage M13in phospholipids bilayers. Biochim Biophys Acta 2001;1511:224–235.

48. Gao C, Mao S, Low C-HL, Wirsching P, Lerner RA, JandaKD. Making artificial antibodies: a format for phage displayof combinatorial heterodynamic arrays. Proc Natl Acad SciUSA 1999; 96:6025–6030.

49. Gao C, Mao S, Ditzel HJ, Farnaes L, Wirsching P, Lerner RA,Janda KD. A cell-penetrating peptide from a novel pVII-pIXphage-displayed random peptide library. Bioorg Med Chem2002; 10:4057–4065.

50. Fransen M, Van Veldhoven PP, Subramani S. Identification ofperoxisomal proteins by usingM13 phage protein VI phage dis-play:molecular evidence thatmammalianperoxisomes containa 2,4-dienoyl-CoA reductase. Biochem J 1999; 340:561–568.

51. Gray CW, Brown RS, Marvin DA. Adsorption complex of fila-mentous fd virus. J Mol Biol 1981; 146:621–627.

52. Rakonjac J, Feng J-N, Model P. Filamentous phage arereleased from the bacterial membrane by a two-step mechan-ism involving a short C-terminal fragment of pIII. J Mol Biol1999; 289:1253–1265.

53. Holliger P, Riechmann L. A conserved infection pathway forfilamentous bacteriophages is suggested by the structure ofthe membrane penetration domain of the minor coat proteing3p from phage fd. Structure 1997; 5:265–275.


54. Lubkowski J, Hennecke F, Pluckthun A, Wlodawer A. Thestructural basis of phage display elucidated by the crystalstructure of the N-terminal domains of g3p. Nat Struct Biol1998; 5:140–147.

55. Smith GP. Filamentous fusion phage:novel expression vec-tors that display cloned antigens on the virion surface.Science 1985; 228:1315–1317.

56. Scott JK, Smith GP. Searching for peptide ligands with anepitope library. Science 1990; 249:386–390.

57. Cross TA, Tsang P, Opella SJ. Comparison of protein anddeoxyribonucleic acid backbone structures in fd and Pf1 bac-teriophages. Biochemistry 1983; 22:721–726.

58. Dotto GP, Enea V, Zinder ND. Functional analysis of the bac-teriophage f1 intergenic region. Virology 1981; 114:463–473.

59. Bauer M, Smith GP. Filamentous phage morphogenetic sig-nal sequence and orientation of DNA in the virion and geneV protein complex. Virology 1988; 167:166–175.

60. Webster RE, Grant RA, Hamilton LAW. Orientation of theDNA in the filamentous bacteriophage f1. J Mol Biol 1981;152:357–374.

61. Shen C, Ikoku A, Hearst JE. Specific DNA orientation inthe filamentous bacteriophage fd as probed by psoralencrosslinking and electron microscopy. J Mol Biol 1979;127:163–175.

62. Webster RE. Filamentous phage biology. In: Barbas CF III,Burton DR, Scott JK, Silverman GJ, eds. Phage Display: ALaboratory Manual. NY: Cold Spring Harbor LaboratoryPress, 2001:1.1–1.37.

63. Fulford W, Model P. J Mol Biol 1988; 203:39–48.

64. Gray CW. Three-dimensional structure of complexes ofsingle-stranded DNA-binding proteins with DNA. IKe andfd gene 5 proteins form left-handed helices with single-stranded DNA. J Mol Biol 1989; 208:57–64.

65. Model P, McGill C, Mazur B, Fulford WD. The replication ofbacteriophage f1: gene V protein regulates the synthesis ofgene II protein. Cell 1982; 29:329–335.

52 Rodi et al.

66. Zaman GJR, Kaan AM, Schoenmakers JGG, Konings RNH.Gene V protein-mediated translational regulation of thesynthesis of gene II protein of the filamentous bacteriophageM13: a dispensable function of the filamentous-phage gen-ome. J Bacteriol 1992; 174:595–600.

67. Wen J-D, Gray CW, Gray DM. SELEX selection of high-affinity oligonucleotides for bacteriophage Ff gene 5 protein.Biochemistry 2001; 40:9300–9310.

68. Torbet J, Gray DM, Gray CW, Marvin DA, Siegrist H. Struc-ture of the fd DNA-gene 5 protein complex in solution—aneutron small-angle scattering study. J Mol Biol 1981;146:305–320.

69. Gray CW, Kneale GG, Leonard KR, Siegrist H, Marvin DA. Anucleoprotein complex in bacteria infected with pf1 filamen-tous phage—identification and electron-microscopic analysis.Virology 1982; 116:40–52.

70. Gray DM, Gray CW, Carlson RD. Neutron-scattering data onreconstituted complexes of fd deoxyribonucleic acid and geneV protein show that the deoxyribonucleic acid is near thecenter. Biochemistry 1982; 21:2702–2713.

71. Manting EH, Driessen AJM. Escherichia coli translocase: theunraveling of a molecular machine. Mol Microbiol 2000;37:226–238.

72. Davis NG, Boeke JD, Model P. Fine structure of a membraneanchor domain. J Mol Biol 1985; 181:111–121.

73. Haigh NG, Webster RE. The major coat protein of filamen-tous bacteriophage f1 specifically pairs in the bacterial cyto-plasmic membrane. J Mol Biol 1998; 279:19–29.

74. Samuelson JC, Chen M, Jiang F, Moller I, Wiedmann M,Kuhn A, Phillips GJ, Dalbey RE. YidC mediates membraneprotein insertion in bacteria. Nature 2000; 406:637–641.

75. Nelson JC, Jiang F, Yi L, Chen M, de Gier J-W, Kuhn A,Dalbey RE. Function of YidC for the insertion of M13procoat protein in Escherichia coli. J Biol Chem 2001; 276:34847–34852.

76. Eisenhawer M, Cattarinussi S, Kuhn A, Vogel H. Fluores-cence resonance energy transfer shows a close helix–helix


distance in the transmembrane M13 procoat protein.Biochemistry 2001; 40:12321–12328.

77. Yuen CTK, Davidson AR, Deber CM. Role of aromaticresidues at the lipid-water interface in micelle-bound bacter-iophage M13 major coat protein. Biochemistry 2000;39:16155–16162.

78. McDonnell PA, ShonK,KimY,Opella SJ. fd coat protein struc-ture inmembrane environments. JMolBiol 1993; 233:447–463.

79. PapavoineCHM,ChristiaansBEC,FolmerRHA,KoningsRNH,Hilbers CW. Solution structure of theM13major coat protein indetergent micelles: a basis for a model of phage assembly invol-ving specific residues. J Mol Biol 1998; 282:401–419.

80. Raposa MP, Webster RE. The filamentous phage assemblyproteins require the bacterial SecA protein for correct locali-zation to the membrane. J Bacteriol 1993; 175:1856–1859.

81. Guy-Caffey JK, Rapoza MP, Jolley KA, Webster RE. Mem-brane localization and topology of a viral assembly protein.J Bacteriol 1992; 174:2460–2465.

82. Horabin JL, Webster RE. An amino acid sequence whichdirects membrane insertion causes loss of membrane poten-tial. J Biol Chem 1992; 263:11575–11583.

83. Guy-Caffey JK, Webster RE. The membrane domain of abacteriophage assembly protein: membrane insertion andgrowth inhibition. J Biol Chem 1993; 268:5496–5503.

84. Raposa MP, Webster RE. The products of gene I and the over-lapping in-frame gene XI are required for filamentous phageassembly. J Mol Biol 1995; 248:627–638.

85. Brissette JL, Russel M. Secretion and membrane integrationof a filamentous phage-encoded morphogenetic protein. J MolBiol 1990; 211:565–580.

86. Linderoth NA, Simon MN, Russel M. The filamentous phagepIV multimer visualized by scanning transmission electronmicroscopy. Science 1997; 278:1635–1638.

87. Russel M. Macromolecular assembly and secretion across thebacterial envelope: type II protein secretion systems. J MolBiol 1998; 279:485–499.

54 Rodi et al.

88. Marciano DK, Russel M, Simon SM. Assembling filamentousphage occlude pIV channels. Proc Natl Acad Sci USA 2001;98:9359–9364.

89. Russel M. Mutants at conserved positions in gene IV, a generequired for assembly and secretion of filamentous phages.Mol Microbiol 1994; 14:357–369.

90. Russel M, Model, P. The role of thioredoxin in filamentousphage assembly. Construction, isolation, and characteriza-tion of mutant thioredoxins. J Biol Chem 1986; 261:14997–15005.

91. Tabor S, Huber HE, Richardson CC. Escherichia coli thiore-doxin confers processivity on the DNA polymerase activityof the gene 5 protein of bacteriophage T7�. J Biol Chem1987; 262:16212–16233.

92. Russel M. Moving through the membrane with filamentousphages. Trends Microbiol 1995; 3:223–228.

93. Russel M. Interchangeability of related proteins and auton-omy of function. The morphogenetic proteins of filamentousphage f1 and IKe cannot replace one another. J Mol Biol1992; 227:453–462.

94. Cross TA, Opella SJ. Protein structure by solid state nuclearmagnetic resonance. Residues 40 to 45 of bacteriophage fdcoat protein. J Mol Biol 1985; 182:367–381.

95. Stark W, Glucksman MJ, Makowski L. Conformation of thecoat protein of filamentous bacteriophage Pf1 determined byneutron diffraction from magnetically oriented gels of specifi-cally deuterated virions. J Mol Biol 1988; 199:171–182.

96. Overman S, Thomas G Jr. Raman spectroscopy of the fila-mentous virus Ff (fd, f1, M13): structural interpretation forcoat protein aromatics. Biochemistry 1995; 34:5440–5451.

97. Overman SA, Tsuboi M, Thomas GJ Jr. Subunit orientationin the filamentous virus Ff (fd, f1, M13). J Mol Biol 1996;259:331–336.

98. Henry GD, Sykes BD. Assignments of the amide 1H and 15NNMR resonances in detergent-solubilized M13 coat protein: amodel for the coat protein dimer. Biochemistry 1992;31:5284–5297.


99. Van de Ven FJM, van Os JWM, Aelen JMA, Wymenga SS,Remerowski ML, Konings RNH, Hilbers CW. Assignment of1H, 15N, and backbone 13C resonances in detergent-solubilizedM13 coat protein via multinuclear multidimensional NMR: amodel for the coat protein monomer. Biochemistry 1993;32:8322–8328.

100. Nambudripad R, Stark W, Opella S, Makowski L. Membrane-mediated assembly of filamentous bacteriophage Pf1 coatprotein. Science 1991; 252:1305–1308.

101. Williams KA, Glibowicka M, Li Z, Li H, Khan AM, ChenYMY, Wang J, Marvin DA, Deber CM. Packing of coat pro-tein amphipathic and transmembrane helices in filamentousbacteriophage M13: role of small residues in protein oligo-merization. J Mol Biol 1995; 252:6–14.

102. Gailus V, Rasched I. The adsorption protein of bacteriophagefd and its neighbor minor coat protein build a structuralentity. Eur J Biochem 1994; 222:927–931.

103. Rakonjac J, Model P. The roles of pIII in filamentous phageassembly. J Mol Biol 1998; 282:25–41.

104. Nelson FK, Friedman SM, Smith GP. Filamentous phageDNA cloning vectors—a non-infective mutant with a non-polar deletion in gene III. Virology 1981; 108:338–350.

105. Crissman JW, Smith GP. Gene-III protein of filamentousphages: evidence for a carboxyl-terminal domain with a rolein morphogenesis. Virology 1984; 132:445–455.

106. Makowski L, Russel M. Structure and assembly of filamen-tous bacteriophages. In: Chiu W, Burnnet RM, Garceia R,eds. Structural Biology of Viruses. Oxford University Press,1992:352–380.

107. Wickner W. Asymmetric orientation of a phage coat proteinin cytoplasmic membrane of Escherichia coli. Proc Natl AcadSci USA 1975; 72:4749–4753.

108. Ohkawa I, Webster RE. The orientation of the major coat pro-tein of bacteriophage f1 in the cytoplasmic membrane ofEscherichia coli. J Biol Chem 1981; 256:9951–9958.

109. Paranchych W. Attachment, ejection and penetration stagesof the RNA phage infectious process. In: Zinder ND, ed.

56 Rodi et al.

RNA Phages. Cold Spring Harbor, NY: Cold Spring HarborLaboratory Press, 1975:85–111.

110. Manchak J, Anthony KG, Frost LS. Mutational analysis ofF-pilin reveals domains for pilus assembly, phage infectionand DNA transfer. Mol Microbiol 2002; 43:195–205.

111. Paranchych W, Frost LS. The physiology and biochemistry ofpili. Adv Microb Physiol 1988; 29:53–114.

112. Stengele I, Bross P, Garces X, Giray J, Rasched I. Dissectionof functional domains in phage fd adsorption protein. Discri-mination between attachment and penetration sites. J MolBiol 1990; 212:143–149.

113. Jacobson A. Role of F-pili in the penetration of bacteriophagef1. J Virol 1972; 10:835–843.

114. Novotny CP, Fives-Taylor P. Retraction of F-pili. J Bacteriol1974; 117:1306–1311.

115. Firth N, Ippen-Ihler K, Skurray R. Structure and function ofthe F factor and mechanism of conjugation. In: Neidhardt FC,ed. Escherichia coli and Salmonella Cellular and MolecularBiology. Washington, DC: American Society of Microbiology,1996:2377–2401.

116. Rondot S, et al. Epitopes fused to F-pilin are incorporatedinto functional recombinant pili. J Mol Biol 1998; 279:589–603.

117. Nagle De Zwaig R, Luria SE. Genetics and physiology ofcolicin-tolerant mutants of Escherichia coli. J Bacteriol1967; 94:1112–1123.

118. Sun T, Webster RE. fii, a bacterial locus required for filamen-tous phage infection and its relation to colicin-tolerant tolAand tolB. J Bacteriol 1986; 165:107–111.

119. Sun T, Webster RE. Nucleotide sequence of a gene clusterinvolved in the entry of the E colicins and the single-strandedDNA of infecting filamentous phage into Escherichia coli. JBacteriol 1987; 169:2667–2674.

120. Click EM, Webster RE. Filamentous phage infection: requiredinteractions with the TolA protein. J Bacteriol 1997; 179:6464–6471.


121. Webster RE. The tol gene products and the import of macro-molecules into Escherichia coli. Mol Microbiol 1991; 5:1005–1011.

122. Braun V, Hermann C. Evolutionary relationship of uptakesystems for biopolymers in Escherichia coli: cross-comple-mentation between the TonB-ExbB-ExbD and the TolA-TolQ-TolR proteins. Mol Microbiol 1993; 8:261–268.

123. Zinder ND. Resistance to colicins E3 and K induced by infec-tion with bacteriophage f1. Proc Natl Acad Sci USA 1973;70:3160–3164.

124. Rampf B, Bross P, Vocke T, Rasched I. Release of periplasmicproteins induced in E coli by expression of an N-terminalproximal segment of the phage fd gene 3 protein. FEBS Lett1991; 280:27–31.

125. Levengood-Freyermuth SK, Click EM, Webster RE. Role of thecarboxyl-terminal domain of TolA in protein import and integ-rity of the outer membrane. J Bacteriol 1993; 175:222–228.

126. Click EM, Webster RE. The TolQRA proteins are required formembrane insertion of the major capsid protein of thefilamentous phage f1 during infection. J Bacteriol 1998;180:1723–1728.

127. Levengood SK, Beyer WF Jr, Webster RE. TolA: a membraneprotein involved in colicin uptake contains an extended heli-cal region. Proc Natl Acad Sci USA 1991; 88:5939–5943.

128. Trenkner E, Bonhoeffer F, Gierer A. The fate of the proteincomponent of bacteriophage fd during infection. BiochemBiophys Res Commun 1967; 28:932–939.

129. Smilowitz H. Bacteriophage f1 infection: fate of the parentalmajor coat protein. J Virol 1974; 13:94–99.

130. Armstrong J, Hewitt JA, Perham RN. Chemical modificationof the coat protein in bacteriophage fd and orientation ofthe virion during assembly and disassembly. EMBO J 1983;2:1641–1646.

131. Glaser-Wuttke G, Keppner J, Rasched I. Pore forming prop-erties of the adsorption protein of filamentous phage fd. Bio-chim Biophys Acta 1989; 985:239–247.

58 Rodi et al.

132. White JM. Membrane fusion. Science 1992; 258:917–924.

133. Krebber C, Spada S, Desplancq D, Krebber A, Liming G,Pluckthun A. Selectively-infective phage (SIP): a mechanisticdissection of a novel in vivo selection for protein-ligand inter-actions. J Mol Biol 1997; 268:607–618.

134. Pratt D, Tzagoloff H, Erdahl WS. Conditional lethal mutantsof the small filamentous coliphage M13. I. Isolation, comple-mentation, cell killing, time of cistron action. Virology 1966;30:397–410.

135. Rodi DJ, Makowski L. Transfer RNA isoacceptor availabilitycontributes to sequence censorship in a library of phage dis-played peptides. Proceedings of the 22nd Tanaguchi Interna-tional Symposium, Seika, Japan, Nov 18–21, 1996.

136. Rodi DJ, Soares A, Makowski L. Quantitative assessment ofpeptide sequence diversity in M13 combinatorial peptidephage display libraries. J Mol Biol 2002; 322(5):1039–1052.

137. Cwirla SE, Peters EA, Barrett RW, Dower WJ. Peptides onphage: a vast library of peptides for identifying ligands. ProcNatl Acad Sci USA 1990; 87:6378–6382.

138. DeGraaf ME, Miceli RM, Mott JE, Fischer HD. Biochemicaldiversity in a phage display library of random decapeptides.Gene 1993; 128:13–17.

139. Marks JD, Hoogenboom HR, Bonnert TP, McCafferty J,Griffiths AD, Winter G. By-passing immunization: humanantibodies from V-gene libraries displayed on phage. J MolBiol 1991; 222:581–597.

140. Christian RB, Zuckermann RN, Kerr JM, Wang L, MalcolmBA. Simplified methods for construction, assessment andrapid screening of peptide libraries in bacteriophage. J MolBiol 1992; 227:711–718.

141. McConnell SJ, Uveges AJ, Fowlkes DM, Spinella DG. Con-struction and screening of M13 phage libraries displayinglong random peptides. Mol Divers 1995; 1:165–176.

142. Noren KA, Noren CJ. Construction of high-complexitycombinatorial phage display peptide libraries. Methods 2001;23:169–178.


143. Ikemura T. Correlation between the abundance of Escheri-chia coli transfer RNAs and the occurrence of the respectivecodons in its protein genes: a proposal for a synonymouscodon choice that is optimal for the E coli translationalsystem. J Mol Biol 1981; 151:389–409.

144. Konigsberg W, Godson GN. Evidence for use of rare codons inthe dnaG gene and other regulatory genes of Escherichia coli.Proc Natl Acad Sci USA 1983; 80:687–691.

145. Zahn K. Overexpression of an mRNA dependent on rarecodons inhibits protein synthesis and cell growth. J Bacteriol1996; 178:2926–2933.

146. Yamane K, Mizushima S. Introduction of basic amino acidresidues after the signal peptide inhibits protein transloca-tion across the cytoplasmic membrane of Escherichia coli.J Biol Chem 1988; 263:19690–19696.

147. Andersson H, von Heijne G. A 30-residue-long ‘‘export initia-tion domain’’ adjacent to the signal sequence is critical forprotein translocation across the inner membrane of Escheri-chia coli. Proc Natl Acad Sci USA 1991; 88:9751–9754.

148. Schuenemann TA, Delgado-Nixon VM, Dalbey RE. Directevidence that the proton motive force inhibits membranetranslocation of positively charged residues within mem-brane proteins. J Biol Chem 1999; 274:6855–6864.

149. Peters EA, Schatz PJ, Johnson SS, Dower WJ. Membraneinsertion defects caused by positive charges in the earlymature region of protein pIII of filamentous phage fdcan be corrected by prlA suppressors. J Bacteriol 1994;176:4296–4305.

150. Duong F, Wickner W. The PrlA and PrlG phenotypes arecaused by a loosened association among the translocaseSecYEG subunits. EMBO J 1999; 18:3263–3270.

151. Randall LL, Hardy SJS. Correlation of competence for exportwith lack of tertiary structure of the mature species: a studyin vivo of maltose-binding protein in E Coli. Cell 1986;46:921–928.

152. Arkowitz RA, Joly JC,WicknerW. Translocation can drive theunfolding of a preprotein domain. EMBO J 1993; 12:243–253.

60 Rodi et al.

153. Von Heijne G. A new method for predicting signal sequencecleavage sites. Nucleic Acids Res 1986; 14:4683–4690.

154. Pluckthun A, Knowles JR. The consequences of stepwisedeletions from the signal-processing site of beta-lactamase.J Biol Chem 1987; 262:3951–3957.

155. Barkocy-Gallagher GA, Cannon JG, Bassford PJ Jr. Beta-turn formation in the processing region is important for effi-cient maturation of Escherichia coli maltose-binding proteinby signal peptidase I in vivo. J Biol Chem 1994; 269:13609–13613.

156. Nilsson I, von Heijne G. A signal peptide with a proline nextto the cleavage site inhibits leader peptidase when present ina sec-independent protein. FEBS Lett 1992; 299:243–246.

157. Kay BK, Adey NB, He Y-S, Manfredi JP, Mataragnon AH,Fowlkes DM. An M13 library displaying 38-amino acid pep-tides as a source of novel sequences with affinity to selectedtargets. Gene 1993; 128:59–65.

158. Ilyichev AA, Minenkova OO, Tatkov SI, Karpyshev NN,Eroshkin AM, Petrenko VA, Sandakhchiev LS. Productionof the M13 phage viable variant with a foreign peptideinserted into the coat basic-protein. Dokl Akad Nauk USSR1989; 307:481–483.

159. Greenwood J, Willis AE, Perham RN. Multiple display offoreign peptides on a filamentous bacteriophage: peptidesfrom Plasmodium falciparum circumsporozoite protein asantigens. J Mol Biol 1991; 220:821–827.

160. Makowski L. Structural constraints on the display of foreignpeptides on filamentous bacteriophages. Gene 1993; 128:5–11.

161. Makowski L, Soares A. Estimating the diversity of peptidepopulations from limited sequence data. Bioinformatics2003; 19:483–489.

162. Lowman HB, Wells JA. Affinity maturation of human growthhormone by monovalent phage display. J Mol Biol 1993;234:564–578.


2

Vectors and Modes of Display

VALERY A. PETRENKO

Department of Pathobiology,College of Veterinary Medicine,

Auburn University, Auburn,Alabama, U.S.A.

GEORGE P. SMITH

Division of Biological Sciences,University of Missouri,

Columbia, Missouri, U.S.A.

I. INTRODUCTION

Construction of molecular chimeras from different sources hasbeen routine in molecular biology since gene splicing began inthe middle of 1970s. For a detailed discussion of the geneticsand biochemistry of molecular cloning systems, refer to thecomprehensive survey of vectors edited by Rodriguez andDenhardt (1). In 1985, recombinant DNA techniques wereused to fashion a new type of chimera that underlies today’sphage display technology (2). To create one of these chimeras,a foreign coding sequence is spliced in-frame into a phage coatprotein gene, so that the ‘‘guest’’ peptide encoded by thatsequence is fused to a coat protein and thereby displayed on

63

the exposed surface of the virion. A phage display library is anensemble of up to about 10 billion such phage clones, eachharboring a different foreign coding sequence, and thereforedisplaying a different guest peptide on the virion surface.The foreign coding sequence can derive from a natural source,or it can be deliberately designed and synthesized chemically.For instance, phage libraries displaying billions of randompeptides can be readily constructed by splicing degeneratesynthetic oligonucleotides into the coat protein gene.

Surface exposure of guest peptides underlies affinityselection, a defining aspect of phage display technology. Atarget binding molecule, which we will call generically the‘‘selector,’’ is immobilized on a solid support of some sort(e.g., on a magnetic bead or on the polystyrene surface of anELISA well) and exposed to a phage display library. Phageparticles whose displayed peptides bind the selector are cap-tured on the support, and can remain there while all otherphages are washed away. The captured phage—generally aminuscule fraction of the initial phage population—can thenbe eluted from the support without destroying phage infectiv-ity, and propagated or cloned by infecting fresh bacterial hostcells. A single round of affinity selection is able to enrich forselector-binding clones by many orders of magnitude; a fewrounds suffice to survey a library with billions or eventrillions of initial clones for exceedingly rare guest peptideswith particularly high affinity for the selector. After severalrounds of affinity selection, individual phage clones are propa-gated and their ability to bind the selector confirmed.

Affinity selection represents a sort of in vitro evolution:the phage library is analogous to a natural population of‘‘organisms’’; affinity for the selector is an artificial analogueto the ‘‘fitness’’ that governs an individual’s survival in the nextgeneration. By introducing ongoing mutation into the ‘‘evol-ving’’ phage population, the analogy with natural selectioncan be made even closer.

General principles and numerous applications of phagedisplay technology are covered in this book and summarizedin a recent review (2). Here, we will focus specifically on thedevelopment and logic of phage display vectors.

64 Petrenko and Smith

II. MOST DISPLAY VECTORS ARE BASEDON FILAMENTOUS PHAGE

Although useful display systems based on bacteriophagesT4, T7, and l have been introduced (3–5), the technology ismost fully developed in the Ff class of filamentous phage,which includes three wild-type strains: f1, first isolated inNew York City (6); M13, isolated in Munich, Germany (7);and fd, isolated in Heidelberg, Germany (8).

These phages are flexible, thread-like particles approxi-mately 1 mm long and 6 nm in diameter. The bulk of their pro-tective tubular capsid (outer shell) consists of 2700 identicalsubunits of the 50-residue major coat protein pVIII arrangedin a helical array with a fivefold rotational axis and a coinci-dent twofold screw axis with a pitch of 3.2 nm; the major coatprotein constitutes 87% of total virion mass (9). Each pVIIIsubunit is largely a-helical and rod-shaped; its axis lies at ashallow angle to the long axis of the virion. About half of its50 amino acids are exposed to the solvent, the other half beingburied in the capsid (for review of phage structure, see Ref.10). At the leading tip of the particle—the end that emergesfirst from the cell during phage assembly (see below)—theouter tube is capped with five copies each of minor coat pro-teins pVII and pIX (encoded by genes VII and IX); five copieseach of minor coat proteins pIII and pVI (encoded by genes IIIand VI) cap the trailing end; it is assumed, but not proven,that the minor proteins form rings that match the fivefoldrotational symmetry of the pVIII array. The capsid enclosesa single-stranded DNA (ssDNA)—the viral or plus strand,whose length is 6407–6408 nucleotides in wild-type strains,but is not constrained by the geometry of the helical capsid.Longer or shorter plus strands—including recombinantgenomes with foreign DNA inserts—can be accommodated ina capsid whose length matches the length of the enclosedDNA by including proportionally fewer or more pVIII subunits.

The wild-type genome is very compact, consisting of 11genes: the five above-mentioned coat protein genes, and sixgenes for proteins involved in viral replication and assembly(Fig. 1). An intergenic region of 508 bases encompasses a

Vectors and Modes of Display 65

packaging (morphogenesis) signal sequence PS and the ori-gins of replication for the minus and plus strands (see Fig. 2for more details).

The phage infects strains of Escherichia coli that harborthe conjugative F episome, which encodes a thread-like appen-dage called the F pilus. As summarized below and detailed inChapter 1, the pilus mediates the infection process, whichculminates in penetration of the plus strand into the cell. Acomplementary ssDNA strand (the minus strand) is thensynthesized by host polymerases to form the double-strandedreplicative form (RF), as illustrated in Fig. 3. Minus-strandsynthesis is initiated with high efficiency by host RNA poly-merase at a special minus-strand origin in the intergenicregion of the ssDNA (shown schematically in Fig. 2), but canoccur at low efficiency in the absence of the minus-strandorigin. Rolling-circle replication of the double-stranded RF

Figure 1 The genome of the fd bacteriophage. Genes are markedI–XI. Viral single-stranded DNA has 6408 nucleotides (126) whichare numbered clockwise from the unique HindII site located in geneII (represented by 0). IG, intergenic region (see diagram in Fig. 2 formore details); PS, the packaging signal; the arrow indicates thepolarity of the single-stranded viral DNA.


produces progeny plus strands, a process that requires thephage replication protein pII acting at a plus-strand originthat also lies in the intergenic region (Fig. 2). Early in infec-tion, progeny plus strands serve as template for minus-strand

Figure 2 Schematic diagram of the intergenic region (IG) of the Ffgenome. Position numbers are as in Fig. 1, according to (126). Thebracketed letters indicate hypothetical hairpin loops in single-stranded DNA as postulated by van Wezenbeek et al.(127) (reviewedin Ref. 128). The first large hairpin [A] located immediately distal togene IV at positions 5500–5577 contains a rho-dependent transcrip-tion termination signal (discussed in Ref. 127) and the packagingsignal PS. Hairpins [B] and [C] are considered the minus-strandorigin; RNA polymerase recognizes this site and synthesizes theshort RNA (shown as a waved dotted line) that primes synthesis ofthe minus strand (129). Hairpins [D] and [E] constitute domain Aof the plus-strand origin. It contains the site at which pII (productof phage gene II) nicks the plus strand of double-stranded RFDNA (between positions 5780 and 5781 of f1 and M13) to initiaterolling-circle extension of the plus strand. B domain of the plus-strand origin increases the initiation of synthesis but is notabsolutely required.


synthesis; balanced plus- and minus-strand synthesis thusresults in the accumulation of double-stranded RF moleculesto a copy number of 100 or more. Late in infection and inchronically infected cells, however, the phage ssDNA-bindingprotein pV sequesters nearly all progeny plus strands intoa filament-shaped complex. These ssDNAs are extrudedthrough the cell envelope, concomitantly shedding pV andacquiring the virion coat proteins from the inner membraneto emerge as completed virions. Extrusion of progeny virionsdoes not kill the cell; chronically infected cells continue todivide, albeit at a slower rate than uninfected cells; it is theslowing of cell division, not cell lysis, that explains plaqueformation by these phages.

All five coat proteins are incorporated into the nascentvirion from the inner membrane of the cell (for review, seeRef. 10). Both the major coat protein pVIII and the minor coat

Figure 3 Infection cycle of filamentous phage Ff (described inSec. I). (Diagram adapted from Ref. 130.)


protein pIII are synthesized with N-terminal signal peptides,which are cleaved from the mature polypeptides (50 aminoacids for pVIII, 406 for pIII) as they are inserted in the innermembrane. Short hydrophobic membrane-anchoring domainsspan the inner membrane, separating the bulk of the proteinfrom a short cytoplasmic C-terminal segment. The other threecoat proteins (pVI, pVII, and pIX) are also membrane pro-teins, though they are not synthesized with signal peptides.

As shown schematically in Fig. 4, the mature pIII proteinconsists of three distinct domains, D1, D2, and D3, linked byglycine-rich tetra- and penta-peptide repeats L1 and L2; inaddition, the C-terminal membrane-anchoring hydrophobic

Figure 4 Mature forms of fd coat proteins. The N-terminus is tothe left. The hydrophobic domains that span bacterial inner mem-brane during phage assembly are underlined. The charged aminoacids are marked by þ or �. The mature pVIII, pIII, pVI, pVII,and pIX proteins have 50, 406, 112, 33, and 32 amino acids, respec-tively. The first 377 residues of pIII consists of three domains sepa-rated by glycine-rich tandem repeat linkers of GGGS and EGGGS;numbers within the circles indicate the residues assigned to eachdomain. (Reproduced from Ref. 10, with permission of CurrentBiology Ltd.)


segment is present. There are significant interactions bet-ween D1, D2, and D3 that probably lead to a compact, stableorganization of the ring of five pIII subunits at the tip of theparticle (11).

Phage infection is initiated by interaction of the middledomain D2 with the tip of the pilus, as illustrated in Fig. 5.The pilus is then hypothesized to retract, drawing the boundphage to the cellular envelope and allowing the N-terminaldomain D1 to interact with periplasmic protein TolA, the cor-eceptor for phage infection. It is thought that the D1 domainis displaced from the D1=D2 complex during phage binding tothe pilus and thus becomes available for interaction with thecoreceptor (12–14) (for review, see Refs. 10, 15). Interdomainlinker L1, separating domains D1 and D2, also participates inphage infection, probably by interacting with the pilus andassisting D1 domain to bind the TolA coreceptor (16).

Figure 5 Initiation of phage infection. CM, cytoplasmic (inner)membrane; OM, outer membrane; D1–D3 domains correspond toD1–D3 domains in the Marvin model (Fig. 4). (a) Phage binds tothe tip of an F pilus. (b) Domain Dl of pIII binds the TolA receptor.(Adapted from Ref. 13 with permission of Elsevier Science Ltd.)


III. GENERAL CLONING VECTORS BASEDON FILAMENTOUS PHAGE

As DNA sequencing by the chain termination method andoligonucleotide-directed mutagenesis came to the fore in the1970s, researchers familiar with filamentous phage sooncame to realize that they have a special virtue as DNA cloningvectors: they deliver one of the two strands of vector DNA to theresearcher in a very convenient form—the virion. The virionsare particularly easy to purify in high yield, and the single-stranded DNA extracted from them can serve as a templatefor chain termination sequencing and oligonucleotide-directedmutagenesis without interference by the complementarystrand. The intergenic region readily tolerates foreign DNAinserts without interfering with viral functions; and thetubular capsid, unlike spherical capsids, imposes nogeometric constraint on the overall length of the viral DNAit encloses.

In 1977, Messing (17) introduced a family of filamentousphage vectors, the M13mp series, that rapidly came to domi-nate this field. These vectors carry an insert of lac DNA thatspans the promoter–operator region and the first 146 codonsof the lacZ gene, which encode an N-terminal peptide ofb-galactosidase called the a peptide. The a peptide by itselfdoes not exhibit enzyme activity, but it restores enzymaticactivity to a defective b-galactosidase that is missing theN-terminal amino acids as a result of a deletion in the lacZgene (deletion DM15). This phenomenon is known as a com-plementation. Vectors in the M13mp series contain uniquerestriction sites within the a peptide coding region (differentvectors in the series carry different combinations of sites).When foreign inserts are spliced into these restriction sitesthey generally disrupt the a peptide and thus a complemen-tation. Presence or absence of a complementation serves asa convenient cloning indicator. When M13mp vectors with-out a foreign insert are plated on medium containing thechromogenic b-galactosidase substrate X-gal, using a bacter-ial host whose resident lacZ gene carries the DM15 deletion,the plaques are blue because of a complementation, whereas


the lawn of uninfected cells is colorless. In contrast, cloneswhose a peptide is disrupted by a foreign insert form colorlessplaques. The same research group installed this cloning indi-cator system in a family of high copy number plasmid vectors,the pUC series, that are the basis of many plasmid vectors inuse today.

When the plus-strand origin of filamentous phage came tobe functionally analyzed in detail, a puzzle emerged: it turnsout that the lac insert in the M13mp series disrupts domainB of that origin (Fig. 2), yet these vectors replicate. The para-dox was resolved with the discovery that M13mp phage har-bors a compensating gene II mutation, which bypasses therequirement for domain B (18). Elsewhere in the intergenicregion, inserts are tolerated with no apparent effect on phagefunction; for example, as shown in Fig. 2, vector fd88-4 has alarge insert between the packaging signal PS and the minus-strand origin.

Many advantages of filamentous phage vectors can becombined with the convenience of very high copy number ina special kind of plasmid vector called a ‘‘phagemid’’ (reviewedin Refs. 19, 20). As will be detailed later, phagemids are thebasis of many phage display systems. These plasmids havean ampicillin (or other antibiotic) resistance gene and twoorigins of replication: a plasmid replication origin—usuallyderived from the pUC plasmid series—that allows them toreplicate to extremely high copy number in an E. coli host;and a filamentous phage replication origin, which is inactiveuntil the phagemid-bearing cell is infected with a filamentousphage acting as a ‘‘helper’’ (Fig. 6). In a helper-infected cellharboring both phagemid and helper phage genomes, phage-encoded proteins act at the phagemid’s phage origin of repli-cation as well as at the helper’s, leading to production ofsingle-stranded phagemid DNA that can be efficiently pack-aged into virion particles. Therefore, two types of infectivevirions are secreted: particles carrying single-stranded helperphage DNA, and particles carrying single-stranded phagemidDNA. Phagemids secreted as virions in these circumstancesare said to have been ‘‘rescued’’ by the helper phage. When ahelper virion infects a cell, the cell acquires the phage DNA


and produces progeny helper virions—usually in sufficientyield and with sufficient infectivity to give visible plaques onagar medium. In contrast, when a phagemid virion infects acell, the cell acquires the antibiotic resistance conferred by

Figure 6 Phagemid display vector. A ‘‘typical’’ phagemid displayvector contains origins of replication for double- and single-strandedDNA synthesis (plasmid and filamentous phage origins), an antibio-tic resistance gene providing selection of transformed bacteria, and afusion gene under the control of a regulated promoter. If the fusiongene derives from phage gene VIII or III, a signal sequence fusedto the coat protein directs secretion of the coat protein and is subse-quently cleaved by signal peptidase, leaving the coat protein span-ning the inner membrane. Phagemid vector is converted intoinfective phage by superinfection of phagemid-bearing cells withhelper phage, as described in the text.


the phagemid (usually ampicillin resistance) but does notproduce virions—until, of course, the phagemid is rescued bysuperinfection with additional helper phage, as illustrated inFig. 6 .

Commonly used as a helper phage, M13K07 is a modifiedM13 phage with the mutant gene II of M13mp1, a phage ori-gin of replication, an additional origin of replication from plas-mid p15A, and a kanamycin resistance gene (21) (Fig. 7). Thekanamycin resistance gene allows cells harboring both phage-mid and helper to be specifically selected by culturing them inmedium containing both kanamycin and ampicillin. Utilizingthe p15A origin, M13K07 is able to replicate independently ofany phage proteins. This allows it to maintain adequate levels

Figure 7 Structure of helper phage M13K07. M13KO7 is an M13derivative which carries the mutation Met40IIe in gene II, theorigin of replication from p15A and the kanamycin resistance gene(Km) from Tn903 both inserted at the AvaI site within the M13origin of replication (position 5825, see Fig. 2). I–XI: genes of M13phage. Viral single-stranded DNA has about 8700 nucleotides (21)which are numbered clockwise from the unique HindII site locatedin gene II (represented by 0). IG, intergenic region (see diagram inFig. 2 for more details); PS, the packaging signal; the arrow showsthe polarity of the single-stranded viral DNA.


of proteins needed for phage production and to overcome theeffects of competition with the phage origin of replication onthe phagemid. In M13K07, as in the M13mp phage, domainB of the phage origin is disrupted, creating a defective originthat may be less active than the wild-type origin on the pha-gemid. This, plus the high copy number of phagemids, leadsto preferential packaging of phagemid DNA into viral parti-cles. When M13K07 is propagated by itself, however, theM13mp1 pII, having the compensating mutation mentionedearlier in this section, functions well enough at the defectivephage origin to maintain a good yield of helper virions.

IV. CLASSIFICATION OF FILAMENTOUSPHAGE DISPLAY SYSTEMS

Before cataloging filamentous phage display systems in theremainder of this review, it will be useful to introduce anomenclature that succinctly captures some of their mostessential features (22). In type 3, 8, 6, 7, and 9 systems(generically, type n systems), the guest peptide is fused toevery copy of the pIII, pVIII, pVI, pVII, or pIX coat protein,respectively; there is a single phage vector genome thatincludes the recombinant coat protein gene (III, VIII, VI,VII, or IX). So far, the only type n systems to be actuallydeveloped are types 3 and 8. Fig. 8 illustrates types of displaysystems exploiting recombinant gene VIII.

A type 88 vector differs from a type 8 vector in that thephage vector genome harbors two genes VIII. One geneencodes the wild-type pVIII subunit; the other has the cloningsites and encodes the recombinant pVIII with the fused guestpeptide. Type 88 virions are therefore mosaics, their capsidsbeing composed of a mixture of recombinant and wild-typepVIII subunits (along with all the other phage coat proteins).Type 33, 66, 77, and 99 systems (generically, type nn) aredefined analogously.

Type 8þ 8 systems differ from type 88 systems in thatthe two genes VIII are on separate genomes (Fig. 8): therecombinant version is on a phagemid, while the wild-typeversion is on a helper phage (phagemid=helper systems were


described in the previous section). Both the helper andrescued phagemid virions, like the type 88 virions describedin the previous paragraph, have mosaic capsids composed ofa mixture of recombinant and wild-type pVIII molecules.Type 3þ 3, 6þ 6, 7þ 7, and 9þ 9 are like type 8þ 8 systems,except that the phagemid carries an insert-bearing recombi-nant gene III, VI, VII, or IX, respectively. We will call thisclass of system generically type nþn.

Type 3þ 3 phagemids must meet a special designrequirement stemming from the fact that cells making pIIImolecules cannot be superinfected by filamentous phage—including helpers. For this reason, it is imperative thatexpression of the phagemid-borne recombinant gene III becontrollable. Gene III expression is shut off as phagemid-bearing cells are being prepared for helper infection, and thenturned on again so the cell can secrete mosaic virions withboth types of pIII subunit.

Figure 8 Classification of phage display vectors. Eight of the 20theoretically possible types are discussed in the text.


In 8þ 8� (or 3þ 3�) type, the helper phage has no geneVIII (or gene III), as illustrated in Fig. 8. As a consequence,all pVIII (pIII) subunits are recombinant, exactly as in type8 (type 3) vectors (discussed in Sec. IX).

V. PHAGE f1—THE FIRST PHAGE-DISPLAYVECTOR

In his first report demonstrating phage display, Smith (23)engineered a type 3 ‘‘fusion phage’’ (as he called it then)from phage f1 using the unique BamHI site in gene III asthe cloning site. A foreign DNA fragment encoding a 57-amino acid segment of EcoRI endonuclease was ligated intothe BamHI site, and E. coli cells were transformed with theresulting recombinant DNA. The transfected cells releasedprogeny phage particles that displayed the 57-residue guestpeptide on their surface as a new segment within the pIIIprotein.

Wild-type f1 turned out to be of little practical use as atype 3 display vector, however. The fusion phage formedvery small or even invisible plaques on the host strains,were 25 times less infective than wild-type phage, and pro-duced deletion mutant phage lacking the foreign insertunder propagation. A plausible reason for these defectsbecame clear when the D1 and D2 domains of pIII were laterdefined (Sec. I, Fig. 4). By chance, the guest peptide hadbeen introduced into a loop in domain D2 that is bridgedby the Cys188-Cys201 disulfide, 20 amino acids away fromthe domain’s C-terminus. It is not known if this loop isdirectly involved in binding the F pilus, but the fact thatantibody to the guest peptide blocks infection by the fusionphage hints that either it or neighboring amino acids mayparticipate in the infection process. Thus, when a guest pep-tide is placed into the loop, it probably hampers the interac-tion of pIII with the pilus and makes the infection processless efficient. The inserted peptide may well influenceassembly and propagation of the phage as well. Subsequentfilamentous display vectors fuse the foreign peptide to either


the N- or C-terminus of the coat protein, and as a rule do notexhibit the severe defects just described.

VI. LOW DNA COPY NUMBER DISPLAYVECTORS BASED ON FD-TET

The first practical type 3 display vectors (24) were derivedfrom a phage called fd-tet (25) by introducing unique restric-tion sites into gene III. This allowed foreign peptides to befused to the amino terminus of the mature pIII protein, wherethey seemed to have much less effect on viral function thanwhen they disrupted domain D2. Vectors derived from fd-tetwere used to construct the first random peptide and antibodylibraries (26–29) and many other applications (for review, seeRef. 2). Some of the fd-tet-derived vectors are included inTable 1.

The DNA copy number—the number of double-strandedphage RF DNA molecules per cell—is very low in the fd-tetfamily. This is because a 2.8-kbp tetracycline resistance deter-minant disrupts the minus-strand origin (in hairpin (B) on theFig. 2), forcing synthesis of this strand to occur through theinefficient alternative pathway. Plaques are so small it isimpractical to quantify infection in terms of plaque-formingunits (pfu). However, because infection transduces the infectedcell to tetracycline resistance, infectious units can be effec-tively quantified as transducing units (TU) by spreading in-fected cells on tetracycline-containing nutrient agar. Thesephages can be propagated independently of infection, even inan uninfectable F� host, by culturing the phage-bearing cellsin medium containing the antibiotic.

In some respects, the replication defect of the fd-tetfamily is an advantage for phage display because it largelyaverts a complication called ‘‘cell killing.’’ When phage assem-bly is fully or partially blocked, intracellular phage DNA andgene products accumulate to toxic levels, and the host cell iskilled without releasing progeny phage (30). Cell killing is lar-gely averted in fd-tet because of its low RF copy number; evensevere morphogenetic defects are readily tolerated (31).


Display vectors based on this phage therefore accommodaterecombinant coat proteins that impair phage assembly, orthat are directly toxic in their own right.

This advantage of fd-tet-based vectors is offset by signi-ficant limitations. Their infectivity—the ratio of successfullyinfected cells to the input of physical particles—is only about0.05 TU=virion, compared to about 0.5 pfu=virion for wild-type

Table 1 Type 3 Vectors

VectorAntibiotic

resistanceCloning

sites Reference

Low DNA copy number vectorsfAFF1 Tetracycline BstXI (27)fd-SfiI=NotI Tetracycline SfiI, NcoI,

XhoI, NotI(71)

fd-tet-DOG1 Tetracycline ApaLI, PstI,SacI, XhoI,NotI

(98)

fd-tetGIIID Tetracycline ApaLI, BbsI,BglII, NotI

(99)

fdTET=Sfi=Not Tetracycline SfiI, NcoI,XhoI, NotI

(100)

fUSE1 Tetracycline PvuII (24)fUSE2 Tetracycline BglII (24)fUSE5 Tetracycline SfiI (26)

High DNA copy number vectorsWild-type f1 None BamHI (23,101)CYT-VI None XhoI, XbaI (102)fd-CAT1 PstI, XhoI (28)m655 Tetracycline XhoI, XbaI (103)m666 None XhoI, XbaI (103)M13KBst Kanamycin BstXI (104)M13KBstX Kanamycin (105)M13KE None KpnI=Acc65I,

EagI(33)

M13LP67 Ampicillin EagI, KpnI (106)M13-PL6 Kanamycin KpnI, BstXI (107)M13stufferbb Ampicillin SacI, KpnI (108)MAEX Ampicillin EagI, XbaI,

NarI(109)

MKTN Kanamycin AccIII, CelII (109)


phage. The 10-fold reduction in infectivity is a result of thedefect in minus-strand synthesis, which is required forconversion of the incoming viral single-stranded DNA todouble-stranded RF DNA, and thus for expression of the tetra-cycline resistance gene. In the interval—typically about45 min—between infection and challenge with tetracycline,the incoming ssDNA has been successfully converted to RFin only a small proportion of the cells. The replication defectin fd-tet also results in a fourfold decrease in particle yield incomparison with the wild-type yield of approximately 2� 1012

virions=mL. The overall yield of infectious units is thereforereduced by a factor of approximately 40-fold. There are alsoindications that fd-tet-derived vectors are genetically unstablewhen propagated in the absence of tetracycline, resulting inclones lacking the tetracycline resistance marker (32).

A number of type 3 vectors that do not have this replica-tion defect have been developed in subsequent years.x

VII. DIVERSITY OF TYPE 3 VECTORS

Type 3 phage vectors differ chiefly with regard to the DNAcopy number (low copy number vectors derived from fd-tet,or high copy number vectors based on wild-type phage orM13mp), antibiotic resistance marker, and cloning sites, assummarized in Table 1 and previously reviewed (2). Includingan antibiotic resistance gene allows isolation of a phage whoseinfectivity is too poor to permit formation of visible plaques—either because of a defect in replication (see the description offd-tet in the previous section) or because the fused foreignpeptide partially impairs pIII function. Such phage clonescan be revealed as colonies of infected (or transfected)bacteria on agar medium containing the antibiotic. However,for display of relatively small peptides (12-mers, for example)in M13mp-derived type 3 vectors, this precaution is not neces-sary; even small plaques can be easily identified by their blueappearance on X-gal-containing media (33). Unique cloningsites allow directional in-frame splicing of foreign peptidessomewhere between the signal peptide and the maturesequence of the coat protein.


Guest peptides displayed on all five pIII subunits areconstrained to lie very close to each other in a ring at onetip of the virion, but their attachment to the virion surfaceis probably quite flexible. For these reasons, it is likely thatsuch displayed peptides can form multivalent interactionswith immobilized selector during affinity selection—bothmultivalent selectors like IgG antibody molecules, and mono-valent selectors arrayed on the solid support at high surfacedensity. Multivalent binding leads to an ‘‘avidity effect’’: avast increase in overall affinity resulting from summation oftwo or more monovalent interactions that are individuallyweak. The avidity effect is an advantage in some applications,but in others, it may undermine selection for peptide ligandswith high monovalent affinity for a target receptor.

VIII. TYPE 8 VECTORS: FIRST LESSONS

Foreign peptides were displayed on pVIII soon after pIII dis-play was introduced (34). The first type 8 constructs weremotivated by the need for polyvalent components in antiviralvaccines and immunodiagnostics; the guest peptide was thusan epitope recognized by an antibody. These constructsdisplayed the guest peptide on every pVIII subunit (35,36),increasing the virion’s total mass by 10%. Yet, remarkably,such particles could retain their ability to infect E. coli andform phage progeny. Such particles were eventually giventhe name ‘‘landscape’’ phage to emphasize the dramaticchange in surface architecture caused by arraying thousandsof copies of the guest peptide in a dense, repeating patternaround the tubular capsid (37–41).

Not surprisingly, phages in which every pVIII subunitbears a guest peptide are defective to some degree. Somerecombinant precoat proteins cannot be processed normallyat the inner membrane of the E. coli cells (42,43), and it islikely that other types of defect are operative as well. In afew cases, this problem can be countered by including an anti-biotic resistance gene in the vector so that clones can beisolated without the need for plaque formation (previous


section); this was the design of the first type 8 display vector,M13B, which included a b-lactamase gene (34,35). Still, mostguest peptides—even small ones—are not tolerated whendisplayed on every pVIII subunit in a high copy number vec-tor, and it was not until the introduction of vectors based onthe replication-defective phage fd-tet (Sec. V) that large land-scape libraries, with 109 or more clones, could be successfullyconstructed (37).

Foreign peptides in landscape phage can subtend asmuch as 25–30% of the virion surface, dramatically changingthe particle’s surface architecture and properties. Dependingupon the particular foreign peptide sequence, the landscapephage can bind organic ligands, proteins, antibodies, or cellreceptors (37–41,44,45), interact with proteases (46), inducespecific immune responses in animals (35,47–49), resist stressfactors such as chloroform or high temperature (37), or mig-rate differently in an electrophoretic gel (Petrenko, unpub-lished). The avidity effect can be even more pronounced inthis display system than in the pIII display systems discussedin the previous section.

Table 2 summarizes type 8 display vectors. The mostadvanced of them—f8-5 and f8-6—allow for insertion of aforeign peptide into any exposed site in pVIII using uniquerestriction sites PstI, BamHI, NheI, and MluI (40). Vectorf8-6 features two TAG amber stop codons between the cloningsites, which prevent contamination of recombinant phageswith the wild-type vector phages. The vector is propagated inan amber-suppressing E. coli strain, while the progeny recom-binant phage clones are grown in a nonsuppressor strain,blocking production of non-insert-bearing vector phages.

Even in a replication-defective vector, there is a stringentlimit to the size and composition of peptides that can be toler-ated on every pVIII subunit (34,36,50,51). This is not becauseof any blanket restriction on incorporation of large recombi-nant pVIII subunits into the virion, however. Using type 88and 8þ 8 vector systems, guest peptides spanning hundredsof amino acids can be displayed on chimeric particles thatinclude wild-type as well as recombinant pVIII subunits[exemplified in our review (2)], as will be discussed in Secs.


VIII and IX. Rather, there seems to be some particular stage ofvirion assembly that cannot accommodate large guest pep-tides. Initiation is an obvious candidate for this hypotheticalspecial stage: perhaps large recombinant subunits cannotform functionally active complexes with the minor coat pro-teins pVII=pIX during initiation of phage assembly (52). The‘‘ban’’ against inserting longer guest peptides into every pVIIIsubunit can sometimes be overcome by randomizing pVIIIamino acids not involved in the phage assembly (37,40,53)(see Sec. XI).

IX. MOSAIC DISPLAY IN TYPE NN SYSTEMS

Mosaic display using type 33 and 88 systems overcomes twopotential disadvantages of type 3 and 8 systems. First, muchlarger, more complex guest peptides can be successfully dis-played on pIII or pVIII when wild-type versions of the coat pro-teins are also available. This is especially so of pVIII display:type 8 vectors cannot accommodate guest peptides longer thanabout 10 amino acids, whereas type 88 vectors can display for-eign peptides containing hundreds of amino acids. Second, theavidity effect is lessened because fewer copies of the guest pep-tides are displayed on the virion surface.

The phage genes are tightly packed into two transcrip-tional domains, as illustrated in Fig. 1. There are accord-ingly only two noncoding segments of the genome availableto accommodate extra genes. The first noncoding area isthe intergenic region, which lies between genes IV and II,and which contains the morphogenesis (packaging) signalPS and the plus and minus origins of DNA replication(Fig. 2). A large tetracycline resistance determinant inter-rupts the minus-strand origin in fd-tet (Sec. V), while a largelac insert disrupts the plus-strand origin in M13mp vectors(Sec. VI). Large inserts can also be accommodated betweenthe packaging signal PS and the minus origin of replication,apparently without compromising any phage functions (seebelow). The second noncoding segment lies between genesVIII and III, and has also been used to accommodate extragenes (54,55).


Table 2 Type 8 Vectors

NameParent

phage

DNAcopynumber

Antibioticresistance Applications Reference

f8-1 fd-tet Low Tetracycline Billion-clone 8-merpeptide library

(37)

f8-5 fd-tet Low Tetracycline Hundredmillion-clonea-helical peptidelibrary

(40)

fdAMPLAY8 fd High Ampicillin Cloning ofpeptides

(43)

fdH fd High None Cloning of4- and 6-merpeptides

(36)

fdISPLAY fd High None Cloning ofpeptides

(110)

M13B M13mp10 High Ampicillin Cloning of5-mer peptides

(34,35)

PM48 f1 High None Ten million-clone8-mer peptidelibrary; small9-mer library

(38,51)

PM54 fd-tet Low Tetracycline Small 6–16-merpeptide libraries

(51)

PM52 fd-tet Low Tetracycline Small 6–16-merpeptide libraries

(51)

84

Petren

koan

dSm

ith

Notes: Amber TAG stop codons in the vectors PM48 and f8-6 are underlined; the signal peptidase cleaves the precoat proteins betweenA-1 and A1.

Vecto

rsan

dModes

ofDisp

lay85

The type 88 vectors of the MB series (56) contain a syn-thetic recombinant gene VIII inserted into the intergenicregion of M13mp18 (replacing the lacZ0 gene and multiplecloning site). The recombinant gene comprises a strong lac(or tac) promoter, a symmetrical lac operator, the signal pep-tide coding sequence from phoA, cloning sites for the insertionof a guest peptide coding sequence, a synthetic codingsequence for the mature form of pVIII, and a strong transcrip-tional terminator. To minimize recombination between therecombinant and wild-type VIII genes, the sequence of therecombinant gene is designed to be very different from thewild-type VIII gene, while specifying the same amino acidsequence. MB virions are mosaics with both wild-type andguest peptide-bearing recombinant pVIII subunits, the pro-portion of which can be regulated by inducing the recombinantgene VIII. In contrast to type 8 vectors, which can accommo-date only relatively short peptides, functional proteins up toapproximately 20 kDa have been displayed on the MB phagesas fusions to the recombinant pVIII protein (57).

Similarly, the JC-M13-88 display vector (58) was con-structed by modifying M13mp18. The a peptide encodingregion of the lacZ0 gene was replaced with the ompA leaderand a synthetic gene VIII cassette (from vector f88-4; see nextparagraph), which are expressed from a tac promoter asfusion with foreign peptides. Under induction of the tac pro-moter with IPTG, the phage gains an average of 1–4 guestpeptide-bearing pVIII subunits per virion, 50% of phage par-ticles being without guest peptides. Other examples of M13-derived type 88 vectors are listed in Table 3 [see also morecomprehensive catalog of vectors in our review (2)].

The type 88 vector f88-4 is derived from the low copynumber phage fd-tet (59). In addition to the wild-type geneVIII, it harbors an artificial gene VIII in the minus origin, nextto the tetracycline resistance gene (Fig. 9). This gene has anIPTG-regulated tac promoter and unique PstI and HindIIIsites for directional cloning of foreign DNA and expressionas a fusion between the signal peptide and the mature pVIII.The vector has been extensively applied for preparation ofnumerous random and natural peptide libraries (59–62).


Guest peptides are likely to be very stable in low copynumber type 88 vectors. To investigate this supposition, anf88-4 clone displaying the FLAG epitope [DYKDDDDKL(63)] was subjected to six rounds of propagation without selec-tion; then about 50 individual subclones were tested for reten-tion of the FLAG epitope. In two independent repetitions of theexperiment, all subclones retained the epitope. In contrast,when the same experiment was carried out with the high copynumber vector fd88-4, which carries the recombinant VIIIexpression cassette between the packaging signal PS and theminus origin of replication (Fig. 2), only 8% and 41% of thesubclones retained the epitope in the two independent repeti-tions. The observed rate of epitope loss in the high copy num-ber vector is undoubtedly low enough to be overcome byselection—even very weak selection. But the rate of loss oflarge guest peptides may well be so high in high copy numbertype 88 vectors as to mandate use of a low copy numberalternative like f88-4.

Table 3 Type nn Vectors

Vector TypeAntibioticresistance Cloning sites Reference

Low DNA copy number vectorsf88-4 88 Tetracycline HindIII, PstI (59)fth1 88 Tetracycline SfiI (55)High DNA copy number vectorsfd88-4 88 None HindIII, PstI (131)fdAMPLAY88 88 Ampicillin SacII, StyI (111)JC-M13-88 88 None XbaI, HindIII (58)MB 88 None a (56)M13IXL604 88 None NcoI, XbaI,

XhoI, SpeI(112,113)

M13702, M13MK100 33 None b (66)pM13Tsn::III 33 None a (65)pM13Tsn::VIII 88 None a (65)

33 (64)

aNote: These vectors were used for display of unique products and have no universalcloning sites.b Data are not available.


Similarly, type 33 vectors (64–66) bear two genes III: onefull-length and one truncated (amino acids 198–408). The for-mer expresses a functional pIII that supports infectivity of thechimeric phage, while the second gene produces a fusionprotein that can be built into the capsid during phageassembly. The type 33 vectors may be genetically unstable,unlike type 88 vectors that use different nucleotide sequences

Figure 9 Vector f88-4. The type 88 vector is derived by splicing a379-bp semi-synthetic major coat protein gene into filamentousphage vector fd-tet. The resulting phage has two major coat proteingenes: the wild-type gene VIII and the artificial, recombinant VIII.The artificial gene is driven by a tac promoter and is therefore indu-cible by IPTG; under fully induced conditions (1 mM IPTG), roughly300 of the 3900 coat protein subunits stem from the recombinantgene VIII, the remainder stemming from the wild-type gene VIII.The recombinant gene VIII has unique HindIII and PstI cloningsites separated by a 21-bp ‘‘stuffer.’’ Replacement of the stuffer withan appropriate synthetic or natural DNA insert can result in displayof up to 300 copies of a peptide encoded by the insert on the virionsurface, fused to the recombinant gene VIII-derived coat proteinsubunits. Genes are marked I–XI. Viral single-stranded DNA has9234 nucleotides which are numbered clockwise from the uniqueHindII site located in gene II (represented by 0). IG, intergenicregion (see diagram in Fig. 2 for more details); PS, the packaging sig-nal; the arrow shows the polarity of the single-stranded viral DNA.


to encode the recombinant and wild-type pVIII polypeptides.There are no reports of type 66, 77, and 99 vectors.

X. MOSAIC DISPLAY IN PHAGEMID SYSTEMS

A ‘‘typical’’ phagemid display vector shown in Fig. 6 containsboth phage and plasmid origins of replication, an antibioticresistance gene providing selection of transformed bacteria,and a fusion gene under the control of a regulated promoter.The signal peptide fused to the coat protein directs secretionof the coat protein and is subsequently cleaved by signal pep-tidase, leaving the coat protein spanning the inner mem-brane. From this position, the protein is then incorporatedinto assembled virions. Some specialized phagemid displayvectors are supplied with additional features—for targetingeukaryotic cells and gene transfer, among other things. Thephagemid pEGFP-lacp8, for example, contains green fluores-cence protein and neo markers that can be used for selectionof transduced eukaryotic cells. It also bears the SV40 origin ofreplication, which allows amplification of plasmid DNA inmammalian cells expressing the SV40 large T antigen (41).

A major advantage of phagemid vectors is that they canbe propagated as a plasmid, when the recombinant coat geneis silent and there is no selection pressure to remove insertedDNA. This contrasts to fusion phages where faster proliferat-ing deletion mutants could quickly dominate. Furthermore,when combining various phagemid and helper partners, thissystem allows for control of display multiplicity from monova-lent to multivalent and can achieve thousands of fusion pro-teins per virion (41). Monovalent display is used for thegeneration of protein phage display libraries and selection ofthe most prominent binders using affinity selection (67,68)(reviewed in Chapter 4). Polyvalent display, in contrast,allows for selection of a broad spectrum of binding candidates.An advanced selection strategy combines the advantages ofboth monovalent and polyvalent display systems, first usingpolyvalent display for selection of primary lead candidates, fol-lowed by monovalent display of the leads to identify the mostavid clones (69–71) (reviewed in Chapter 4).


Monovalent display exploits pIII (71) or its C-terminaldomain [amino acids 198–406 or 230–406 (72)] for fusion ofdisplayed peptides, and provides controlled expression of thepIII fusion, which can be as low as one copy per 100 phageparticles (68). Since N-terminal domains of pIII expressed inE. coli cells hinder infection of the cells with phage (73), dele-tion of these domains allows for superinfection of the cellswith the helper phage. The multiplicity of display can be con-trolled through several different means: (a) using induciblepromoters (74); (b) switching between pVIII (8þ 8) and pIII(3þ 3) display systems (50); (c) mutating the recombinantcoat protein (75); or (d) using ‘‘hyper’’ helper phages with geneIII or VIII deleted. The later type of display may be called3þ 3� or 8þ 8�, emphasizing that the second gene III or VIIIcoming from a helper phage is missing (illustrated in Fig. 8)(41,76–78). If the helper phage is lacking gene VIII, the pha-gemid DNA is coated in the homogenous landscape formatwith the recombinant pVIII (41), exactly like type 8 vectors.The defective helper phages can be produced in cells that har-bor the plasmids containing gene III or VIII under control ofthe tac or phage shock protein (psp) promoters (41,79). Thesalient feature of the psp promoter is that it is activated inthe presence of bacteriophage protein IV, and is thereforesilent until the moment of infection.

Minor coat proteins pVII and pIX have been used inphagemid format for display of heavy and light antibodychains (80). Fusion proteins were expressed from the phage-mid as procoats with ompA and pelB leaders, unlike thenative pVII and pIX that are synthesized without leadersand require a processing step. Since these two proteinsappear to interact with one another in the phage capsid,they may be ideal for the display of dimeric proteins suchas antibodies or integrins. Later, aspects of this technologywere invoked and extended to construct a large, humansingle-chain Fv (scFv) antibody library displayed on pIX(81). The type 6þ 6 format that allows C-terminal display isdescribed separately in Sec. X. Some examples of phagemiddisplay systems are presented in Table 4.


XI. VECTORS FOR C-TERMINAL DISPLAY

Since the commonly used display systems utilize fusion offoreign peptides to the N-terminus of pIII or pVIII, they areunsuitable for surface expression of full-length cDNA bearingstop codons. However, another coat protein, pVI, can tolerateC-terminal fusion to foreign peptides without abolishing phageviability (82). Thus, cDNAs can be ligated to the 30 end of geneVI using phagemid vector pDONG6, allowing fusion of foreigncoding sequences to gene VI in all three possible frames.

Table 4 Phagemid Display Systems

NameRecombinantphage gene

Antibioticresistance Cloning sites Reference

PC89 VIII None EcoRI, BamHI (50)pCANTAB 5E III Ampicillin SfiI, NotI PharmaciapCES1 III Ampicillin ApaLI, NotI (114)pCGMT-1b VII, IX SacI, NcoI (80)pComb3 D3-III Ampicillin XhoI, SpeI (115)pComb3H,pComb3X

D3-III Ampicillin SfiI (72)

pComb3-M3 D3-III Ampicillin Multiple (116)pComb8 VIII Ampicillin (117)pDONG6 VI Ampicillin SfiI, NotI,

BamHI(82)

pEGFP-lacp8 VIII Kanamycin BpuAI BamHI (41)pETT7GSTgp8 VIII Ampicillin SacII, StuI,

XhoI, NotI(118)

pEXmide 3 III Ampicillin Multiple (119)pGEM-gIII D3-III Ampicillin SacI, XhoI,

NheI(120)

pGP-F100 III Ampicillin SfiI (121)pGZ1 III Ampicillin NotI, SfiI (122)phGH-M13gIII D3-III Ampicillin a (123)pHEN1 III Ampicillin SfiI, NotI (98)pSEX III Ampicillin,

chloram-phenicol

BamHI, PstI (124)

pSEX40 III Ampicillin Multiple (125)

aNote: These vectors were used for display of unique products and have no universalcloning sites.


C-terminally fused peptides have also been displayedindirectly on pIII via a leucine zipper ‘‘fastener’’ (83,84), asillustrated in Fig. 10. The phagemid bears two recombinantgenes: the coding sequence for the Jun half of the AP-1 zipper

Figure 10 pJuFo phagemid—a vector for C-terminal display. Thediagram illustrates the proposed pathway for display of cDNA pro-ducts on the phage surface explained in the text. (Adapted from (83)with permission from Elsevier Science-NL.)


fused N-terminally to phage gene III, and the coding sequencefor the Fos half of the zipper fused N-terminally to a cDNAlibrary. Both fusion proteins are preceded by signal peptidesand expressed from IPTG-inducible lac promoters. When cellsharboring the phagemid are superinfected with helper phageand induced with IPTG, the two proteins are secreted into theperiplasm without their signal peptides, where the two‘‘half-zippers’’ join prior to, or concomitantly with, incorpora-tion into the secreted virions.

Recently, a new approach was developed allowingdisplay of foreign proteins on the C-terminus of an artificialcoat protein (53). Normally, as described above, the negativelycharged N-terminus of pVIII is exposed on the surface of thephage while the C-terminus forms a positively chargedlumen. Using a combination of rational design and extensivemutagenesis, Sidhu and colleagues developed an artificialpVIII with a reversed sequence of functionally importantamino acids, which exposes its negatively charged C-terminuson the surface of the capsid. It is unlikely that the proteinitself supports phage assembly, but it can be introduced intoa phage capsid with low probability during phage morphogen-esis supported by an excess of the normal pVIII (85).

XII. PHAGE PROTEINS AS CONSTRAININGSCAFFOLDS

It is commonly accepted that the binding affinities of peptidesmay be higher when they are displayed as segments of foldedproteins, called ‘‘scaffolds.’’ A scaffold displays foreign pep-tides in a specific constrained conformation, which allowsthe peptide to bind a receptor without paying an entropy pen-alty. For example, the natural immunoglobulin scaffold inantibodies brings six antigen-binding peptides together andhelps them to collaborate in binding of an antigen. Suchscaffolds, along with fused random peptides, can be displayedon the phage surface and used for selection of binding entities.At the same time, phage itself may serve as a scaffold for theconstrained presentation of foreign peptides if these peptides


replace exposed segments of the coat proteins without dis-turbing their general architecture and phage viability. Theminor coat protein pIII and the major coat protein pVIII canbe considered as primary candidate scaffolds because theirstructures and functions have been studied in great detail.However, pIII was used only for N-terminal presentation ofpeptides and, in its truncated form, for display of functionaldomains of foreign proteins in their intrinsic conformation,as described above in Sec. VI. Study of the pVIII protein asa constraining scaffold is more advanced.

The conformation of major coat protein pVIII in phagecapsid has been studied using physical methods and con-firmed by mutagenesis (reviewed in Ref. 86). It consists of fourstructural domains A–D, as shown below.

The N-terminal mobile surface domain A (Ala-1 to Asp-5)has a nonhelical, possibly disordered conformation (87).Domain B is an amphipathic, gradually curving a-helix,extending from Pro-6 to about Tyr-24 (88). Domain C, a highlyhydrophobic a-helix extending from Ala-25 to Ala-35, isentirely buried in the interior of the protein coat. The remain-der of protein D, from Thr-36 to Ser-50, constitutes an amphi-pathic a-helix forming the inside wall of the protein coat. Fourbasic residues near the C-terminus interact with the viralDNA. The N-terminus of pVIII is exposed on the outer surfaceand its C-terminus lies in the lumen (10). Neighboring pVIIIsubunits make numerous inter-sub-unit contacts that impartremarkable physical stability to the structure. The length ofthe capsid is directly proportional to the length of the enclosedviral DNA and inversely proportional to the number ofpositively charged residues at the luminal (C-terminal) endof the pVIII polypeptide (89). According to the a-helical modelof segment B, 11 residues (Lys-8, Asp-12, Ser-13, Gln-15,


Ala-16, Ser-17, Thr-19, Glu-20, Tyr-21, Gly-23, and Tyr-24)comprise a polar area exposed on the surface of the phage,while the other residues constitute a nonpolar region thatinteracts with the phage body (86,90,91).

In type 8 systems, where only one kind of pVIII is avail-able, only minor pVIII sequence variants can be toleratedwithout abolishing phage assembly or infectivity, includingpoint mutations (91), deletion of a few N-terminal aminoacids (40), or insertion of very short peptides into the N-ter-minus (34,36–38,46,50,51,92). Much more drastic changes inpVIII can be tolerated in type 8þ 8 systems, however. Inparticular, it is possible to incorporate a few copies of recom-binant pVIII bearing a large foreign domain with hundredsof amino acids into a capsid that is mainly composed ofwild-type pVIII subunits. In fact, some mutations in therecombinant gene VIII even increase incorporation effi-ciency, enabling the development of improved phage displayplatforms (75,93).

In order to explore pVIII as a scaffold, two domains,A and B, were replaced by foreign peptides using type 8 vec-tors described in Sec. VII. Peptides fused to N-terminaldomain A can probably adopt many different conformationsdepending on their sequence, while peptides loaded intodomain B conform to the a-helical architecture of the flankingwild-type residues (40). In this way, it is possible to generatevery large libraries of surface-accessible a-helical ligands thatcan serve as a source of potential drugs, vaccines, and substi-tute antibodies (40).

XIII. CONCLUSION

Table 5 summarizes salient features of the display systemsdescribed in this chapter. For type n and nn vectors, DNA copynumber is an important parameter. High DNA copy numbervectors are convenient because of their high infectivity andphage production. However, inserts are much less stable inthese vectors, and they may not be suitable for long insertssuch as antibody fragments. Insert stability has not been


systematically investigated in nþn phagemid systems, butsince phagemids can have very high DNA copy numbers,insert instability is likely to be a concern in these systems too.

In many display systems, there is no sharp limit on thelength of the displayed peptide. However, type 8 vectors willonly accommodate very short peptides—especially in highDNA copy number systems. In type 3 systems, long insertstend to impair infectivity. In all display systems, the numberof peptides displayed intact on the virion surface tends to bemuch lower for long peptides than for short ones. The poorerdisplay density of long peptides may be due to less efficientincorporation or to more efficient degradation by the potentcytosolic and periplasmic proteases that eliminate malfoldedproteins in E. coli.

Table 5 Summary of Display Systems

Type

DNAcopynumber

Length ofpeptidesaccepted

Number ofpeptidesper virion

Position ofpeptidefusion Special considerations

3 High Unknown 5 N-terminus Long peptides may begenetically unstable

3 Low No limit 5 N-terminus8 High Very short >2,700 N-terminus No cysteines (37)8 Low �10 >2,700 N-terminus No cysteines (37);

highly constraineda-helical scaffold

88 High Unknown �300 N-terminus Long peptides may begenetically unstable

88 Low No limit �300 N-terminus33 Low No limit <1 N-terminus33 High <1 N-terminus3þ3 No limit <1 N-terminus Often used for

monovalent display;C-terminal fusionpossible throughleucine zipperfastener (Sec. X)

8þ8 No limit 100–1000 N-terminus8þ8 Unknown <1 C-terminus See Sec. X6þ6 No limit <1 N-terminus6þ6 No limit <1 C-terminus See Sec. X9þ9 No limit <1 N-terminus


Some display systems are effectively ‘‘monovalent,’’ inthat the number of peptides displayed per virion is one oreven much less. Low display density can be the result of poorincorporation or high degradation (e.g., in C-terminal type6þ 6 or 8þ 8 systems), or purposely fostered by providingwild-type subunits to compete with peptide-bearing subunitsfor incorporation into virions (as in type 3þ 3 systems). Mono-valent display is an important advantage when it is desired toselect for high monovalent affinity. Multivalent displayundermines selection for this property, by greatly increasingavidity (effective affinity) to the point where weak and strongligands cannot be distinguished. In other applications, how-ever, multivalency is a distinct advantage. Such is the casewhen it is desirable to accumulate a broad spectrum of pep-tides as potential leads; the identified peptides can be orderedsubsequently in accordance with their affinity and selectivityusing monovalent display. Polyvalent display vectors are par-ticularly effective for targeting multiple receptors on cell sur-faces, especially for triggering receptor-mediated endocytosis(41,44,94,95). ‘‘Landscape’’ phage using type 8 vectors arean extreme case of multivalency, in which thousands of copiesof the displayed peptide are arranged in a regular array onthe virion surface. This ultra-high density display refashionsa substantial fraction of the surface chemical architecture,and can lead to ‘‘emergent’’ properties of the virion as a wholethat would be hard to foresee from the properties of the indi-vidual peptides. Such virions can be looked on as a new,biologically selectable type of ‘‘nanofiber.’’

In preparing very large peptide or antibody phagelibraries, phagemid vectors may be preferred because theirDNA copy number is much higher than that of phage. Thisallows easy purification of large amounts of high-puritydouble-stranded vector DNA for library construction. On theother hand, the necessity for superinfection with helper phageintroduces a significant complication into their use—a compli-cation that is encountered many thousands of times for agiven library, not just once at the stage of its construction.Expression of the fusion genes in phagemid systems is alsoreported to be less efficient than in phage vectors (71).


Phage display is constantly being applied to new experi-mental and practical problems, motivating the engineering ofvectors with new unique characteristics. For example, devel-opment of advanced gene-delivery systems would requirevectors of small size with enhanced ability for transfectionof eukaryotic cells (41,96). Design of specific probes for detec-tion would require environmentally stable vectors that canselfassemble to generate bioselective layers on biosensors(97). Other applications of phage display technology willdemand even more specialized new vector systems.

REFERENCES

1. Rodriguez RL, Denhardt DT, eds. Vectors: A Survey ofMolecular Cloning Vectors and Their Uses. Stoneham, MA:Butterworth Publishers, 1987.

2. Smith GP, Petrenko VA. Phage display. Chem Rev 1997; 410:97–391.

3. Ansuini H, Cicchini C, Nicosia A, Tripodi M, Cortese R,Luzzago A. Biotin-tagged cDNA expression libraries dis-played on lambda phage: a new tool for the selection ofnatural protein ligands. Nucleic Acids Res 2002; 30:e78.

4. Houshmand H, Froman G, Magnusson G. Use of bacterioph-age T7 displayed peptides for determination of monoclonalantibody specificity and biosensor analysis of the bindingreaction. Anal Biochem 1999; 268:363–370.

5. Malys N, Chang DY, Baumann RG, Xie D, Black LW. A bipar-tite bacteriophage T4 SOC and HOC randomized peptidedisplay library: detection and analysis of phage T4 terminase(gp17) and late sigma factor (gp55) interaction. J Mol Biol2002; 319:289–304.

6. Loeb T. Isolation of bacteriophage specific for the Fþ andHfr mating types of Escherichia coli K12. Science 1960; 131:932–933.

7. Hofschneider PH. Untersuchungen uber ‘‘kleine’’ E.coli K12bacteriophagen M12, M13, und M20. Z Naturforschg 1963;18b:203–205.


8. Marvin DA, Hoffman-Berling H. Physical and chemical prop-erties of two new small bacteriophages. Nature (London)1963; 197:517–518.

9. Berkowitz SA, Day LA. Mass, length, composition and struc-ture of the filamentous bacterial virus fd. J Mol Biol 1976;102:531–547.

10. Marvin DA. Filamentous phage structure, infection andassembly. Curr Opin Struct Biol 1998; 8:150–158.

11. Chatellier J, Hartley O, Griffiths AD, Fersht AR, Winter G,Riechmann L. Interdomain interactions within the gene 3protein of filamentous phage. FEBS Lett 1999; 463:371–374.

12. Holliger P, Riechmann L, Williams RL. Crystal structure ofthe two N-terminal domains of g3p from filamentous phagefd at 1.9 A: evidence for conformational lability. J Mol Biol1999; 288:649–657.

13. Lubkowski J, Hennecke F, Pluckthun A, Wlodawer A. Fila-mentous phage infection: crystal structure of g3p in complexwith its coreceptor, the C-terminal domain of TolA. Structure1999; 7:711–722.

14. Deng LW, Malik P, Perham RN. Interaction of the glo-bular domains of pIII protein of filamentous bacteriophagefd with the F-pilus of Escherichia coli. Virology 1999; 253:271–277.

15. Webster R. Filamentous Phage Biology, in Phage Display. ALaboratory Manual. In: Barbas CF, et al, eds. Cold SpringHarbor, New York: Cold Spring Harbor Laboratory Press,2001:37, 1.1–1.

16. Nilsson N, Malmborg AC, Borrebaeck CAK. The phage infec-tion process: a functional role for the distal linker region ofbacteriophage protein 3. J Virol 2000; 74:4229–4235.

17. Messing J. Cloning single-stranded DNA. Mol Biotechnol1996; 5:39–47.

18. Dotto GP, Zinder ND. Reduction of the minimal sequencefor initiation of DNA synthesis by qualitative or quanti-tative changes of an initiator protein. Nature 1984; 311:279–280.


19. Mead DA, Kemper B. Chimeric single-stranded DNA phage-plasmid cloning vectors. In: Rodriguez RL, Denhardt DT,eds. Vectors: A Survey of Molecular Cloning Vectors andTheir Uses. Stoneham, MA: Butterworth Publishers, 1988:85–102.

20. Cesareni G. Phage-plasmid hybrid vectors. In: Rodriguez RL,Denhardt DT, eds. Vectors: A Survey of Molecular CloningVectors and Their Uses. Stoneham, MA: ButterworthPublishers, 1988:103–111.

21. Vieira J, Messing J. Production of single-stranded plasmidDNA. Methods Enzymol 1987; 153:3–11.

22. Smith GP. Preface. Surface display and peptide libraries.Gene 1993; 128:1–2.

23. Smith GP. Filamentous fusion phage: novel expressionvectors that display cloned antigens on the virion surface.Science 1985; 228:1315–1317.

24. Parmley SF, Smith GP. Antibody-selectable filamentous fdphage vectors: affinity purification of target genes. Gene1988; 73:305–318.

25. Zacher ANI, Stock CA, Golden JWI, Smith GP. A newfilamentous phage cloning vector: fd-tet. Gene 1980; 9:127–140.



28. McCafferty J, Griffiths AD, Winter G, Chiswell DJ. Phageantibodies: filamentous phage displaying antibody variabledomains. Nature 1990; 348:552–554.

29. Clackson T, Hoogenboom HR, Griffiths AD, Winter G. Mak-ing antibody fragments using phage display libraries. Nature1991; 352:624–628.

30. Pratt D, Tzagoloff H, Erdahl WS. Conditional lethal mutantsof the small filamentous coliphage M13. I. Isolation, comple-mentation, cell killing, time of cistron action. Virology 1966;30(3):397–410.


31. Smith GP. Filamentous phage assembly: morphogeneticallydefective mutants that do not kill the host. Virology 1988;167(1):156–165.

32. Enshell-Seijffers D, Smelyanski L, Vardinon N, Yust I,fGershoni JM. Dissection of the humoral immune responsetoward an immunodominant epitope of HIV: a model for theanalysis of antibody diversity in HIV plus individuals.FASEB J 2001; 15:2112–2120.

33. Noren KA, Noren CJ. Construction of high-complexity combi-natorial phage display peptide libraries. Methods 2001; 23:169–178.

34. Ilyichev AA, Minenkova OO, Tatkov SI, Karpyshev NN,Eroshkin AM, Petrenko VA, Sandakhchiev LS. Constructionof M13 viable bacteriophage with the insert of foreign pep-tides into the major coat protein. Dokl Biochem (Proc AcadSci USSR) Engl Transl 1989; 307:196–198.

35. Minenkova OO, Ilyichev AA, Kishchenko GP, Petrenko VA.Design of specific immunogens using filamentous phage asthe carrier. Gene 1993; 128:85–88.

36. Greenwood J, Willis AE, Perham RN. Multiple display offoreign peptides on a filamentous bacteriophage. Peptidesfrom Plasmodium falciparum circumsporozoite protein asantigens. J Mol Biol 1991; 220:821–827.

37. Petrenko VA, Smith GP, Gong X, Quinn T. A library oforganic landscapes on filamentous phage. Protein Eng 1996;9:797–801.

38. Iannolo G, Minenkova O, Gonfloni S, Castagnoli L, CesareniG. Construction, exploitation and evolution of a new pep-tide library displayed at high density by fusion to the majorcoat protein of filamentous phage. Biol Chem 1997; 378:517–521.

39. Petrenko VA, Smith GP. Phages from landscape libraries assubstitute antibodies. Protein Eng 2000; 13:589–592.

40. Petrenko VA, Smith GP, Mazooji MM, Quinn T. Alpha-helically constrained phage display library. Protein Eng2002; 15:943–950.


41. Legendre D, Fastrez J. Construction and exploitationin model experiments of functional selection of a landscape library expressed from a phagemid. Gene 2002; 290:203–215.

42. Malik P, Terry TD, Gowda LR, Langara A, Petukhov SA,Symmons MF, Welsh LC, Marvin DA, Perham RN. Role ofcapsid structure and membrane protein processing in deter-mining the size and copy number of peptides displayed onthe major coat protein of filamentous bacteriophage. J MolBiol 1996; 260:9–21.

43. Malik P, Terry TD, Bellintani F, Perham RN. Factors limit-ing display of foreign peptides on the major coat protein offilamentous bacteriophage capsids and a potential role forleader peptidase. FEBS Lett 1998; 436:263–266.

44. Romanov VI, Durand DB, Petrenko VA. Phage display selec-tion of peptides that affect prostate carcinoma cells attach-ment and invasion. Prostate 2001; 47:239–251.

45. Bishop-Hurley SL, Mounter SA, Laskey J, Morris RO, ElderJ, Roop P, Rouse C, Schmidt FJ, English JT. Phage-displayedpeptides as developmental agonists for Phytophthora capsicizoospores. Appl Environ Microbiol 2002; 68:3315–3320.

46. Terry TD, Malik P, Perham RN. Identification of Hcv coremimotopes—improved methods for the selection and use ofdisease-related phage-displayed peptides. Biol Chem 1997;378:495–502.

47. di Marzo Veronese F, Willis AE, Boyer-Thompson C, AppellaE, Perham RN. Structural mimicry and enhanced immuno-genicity of peptide epitopes displayed on filamentous bacter-iophage. The V3 loop of HIV-1 gp120. J Mol Biol 1994;243:167–172.

48. Perham RN, Terry TD, Willis AE, Greenwood J, di MarzoVeronese F, Appella E. Engineering a peptide epitope displaysystem on filamentous bacteriophage. FEMS Microbiol Rev1995; 17:25–31.

49. De Berardinis P, Sartorius R, Fanutti C, Perham RN, DelPozzo G, Guardiola J. Phage display of peptide epitopes fromHIV-1 elicits strong cytolytic responses. Nat Biotechnol 2000;18:873–876.


50. Felici F, Castagnoli L, Musacchio A, Jappelli R, Cesareni G.Selection of antibody ligands from a large library of oligopep-tides expressed on a multivalent exposition vector. J Mol Biol1991; 222:301–310.


52. Endemann H, Model P. Location of filamentous phage minorcoat proteins in phage and in infected cells. J Mol Biol 1995;250:496–506.

53. Sidhu SS. Engineering M13 for phage display. Biomol Eng2001; 18:57–63.

54. Krebber C, Spada S, Desplancq D, Pluckthun A. Co-selectionof cognate antibody–antigen pairs by selectively-infectivephages. FEBS Lett 1995; 377:227–231.

55. Enshell-Seijffers D, Smelyanski L, Gershoni JM. The rationaldesign of a ‘type 88’ genetically stable peptide display vector inthe filamentous bacteriophage fd. Nucleic Acids Res 2001;29:E50–50.

56. Markland W, Roberts BL, Saxena MJ, Guterman SK, LadnerRC. Design, construction and function of a multicopy displayvector using fusions to the major coat protein of bacteriophageM13. Gene 1991; 109:13–19.

57. Roberts BL, Markland W, Ladner RC. Affinity maturationof proteins displayed on surface of M13 bacteriophage asmajor coat protein fusions. Methods Enzymol 1996; 267:68–82.

58. Chappel JA, He M, Kang AS. Modulation of antibody displayon M13 filamentous phage. J Immunol Methods 1998;221:25–34.

59. Zhong G, Smith GP, Berry J, Brunham RC. Conformationalmimicry of a chlamydial neutralization epitope on filamen-tous phage. J Biol Chem 1994; 269:24183–24188.

60. Kouzmitcheva GA, Petrenko VA, Smith GP. Identifying diag-nostic peptides for Lyme disease through epitope discovery.Clin Diagn Lab Immunol 2001; 8:150–160.


61. Matthews LJ, Davis R, Smith GP. Immunogenically fit subu-nit vaccine components via epitope discovery from naturalpeptide libraries. J Immunol 2002; 169:837–846.

62. Bonnycastle LLC, Shen J, Menendez A, Scott JK. Productionof peptide libraries. In: CF Barbas III, et al., ed. Phage Dis-play. A Laboratory Manual. Cold Spring Harbor, New York:Cold Spring Harbor Laboratory Press 2001:16.11–16.28.

63. Hopp TP, Prickett KS, Price VL, Libby RT, March CJ, CerettiDP, Urdal DLa, Conlon PJ. A short polypeptide markersequence useful for recombinant protein identification andpurification. Biotechnology 1988; 6:1204–1210.

64. Corey DR, Shiau AK, Yang Q, Janowski BA, Craik CS. Tryp-sin display on the surface of bacteriophage. Gene 1993;128:129–134.

65. Wang CI, Yang Q, Craik CS. Phage display of proteases andmacromolecular inhibitors. Methods Enzymol 1996; 267:52–68.

66. McConnell SJ, Kendall ML, Reilly TM, Hoess RH. Con-strained peptide libraries as a tool for finding mimotopes.Gene 1994; 151:115–118.

67. Baas S, Greene R, Wells JA. Hormone phage: an enrichmentmethod for variant proteins with altered binding properties.Proteins 1990; 8:309–314.

68. Lowman HB. Structural and mechanistic determinants ofaffinity and specificity of ligands discovered or engineeredby phage [review]. Annu Rev Biophys Biomol Struct 1997;26:27–45.

69. Wrighton NC, Farrell FX, Chang R, Kashyap AK, BarboneFP, Mulcahy LS, Johnson DL, Barrett RW, Jolliffe LK, DowerWJ. Small peptides as potent mimetics of the proteinhormone erythropoietin. Science 1996; 273:458–464.

70. Yanofsky SD, Baldwin DN, Butler JH, Holden FR, JacobsJW, Balasubramanian P, Chinn JP, Cwirla SE, Peters-BhattE, Whitehorn EA, Tate EH, Akeson A, Bowlin TL, Dower WJ,Barrett RW. High affinity type I interleukin 1 receptorantagonists discovered by screening recombinant peptidelibraries. Proc Natl Acad Sci USA 1996; 93:7381–7386.


71. O’Connell D, Becerril B, Roy-Burman A, Daws M, Marks JD.Phage versus phagemid libraries for generation of humanmonoclonal antibodies. J Mol Biol 2002; 321:49–56.

72. Carlos F, Barbas III, Dennis R, Burdon, Jamie K, Scott,Gregg J. Silverman. Phage-display vectors. In: Carlos F.Barbas III, Dennis R. Buton, Jamie K. Scott, Gregg J. Silver-man, eds. Phage Display: A Laboratory Manual. Cold SpringHarbor, New York: Cold Spring Harbor Laboratory Press,2001:2.1–2.19.

73. Boeke Jd, Model P, Zinder ND. Effects of bacteriophage f1gene III protein on the host cell membrane. Mol Gen Genet1982; 186:185–192.

74. Lutz R, Bujard H. Independent and tight regulation of tran-scriptional units in Escherichia coli via the LacR=O, theTetR=O and AraC=I1-I2 regulatory elements. Nucleic AcidsRes 1997; 25:1203–1210.

75. Sidhu SS, Weiss GA, Wells JA. High copy display of largeproteins on phage for functional selections. J Mol Biol 2000;296:487–495.

76. Griffiths AD, Malmqvist M, Marks JD, Bye JM, EmbletonMJ, McCafferty J, Baier M, Holliger KP, Gorick BD,Hughes-Jones NC, et al. Human anti-self antibodies withhigh specificity from phage display libraries. EMBO J 1993;12:725–734.

77. Duenas M, Borrebaeck CA. Novel helper phage design: inter-genic region affects the assembly of bacteriophages and thesize of antibody libraries. FEMS Microbiol Lett 1995; 125:317–321.

78. Rondot S, Koch J, Breitling F, Dubel S. A helper phage toimprove single-chain antibody presentation in phage display.Nat Biotechnol 2001; 19:75–78.

79. Rakonjac J, Jovanovic G, Model P. Filamentous phageinfection-mediated gene expression: construction and propa-gation of the gIII deletion mutant helper phage R408d3. Gene1997; 198:99–103.

80. Gao C, Mao S, Lo CH, Wirsching P, Lerner RA, Janda KD.Making artificial antibodies: a format for phage display of


combinatorial heterodimeric arrays. Proc Natl Acad Sci USA1999; 96:6025–6030.

81. Gao C, Mao S, Kaufmann G, Wirsching P, Lerner RA, JandaKD. A method for the generation of combinatorial antibodylibraries using pIX phage display. Proc Natl Acad Sci USA2002; 99:12612–12616.

82. Jespers LS, Messens JH, De Keyser A, Eeckhout D, Van denBrande I, Gansemans YG, Lauwereys MJ, Vlasuk GP,Stanssens PE. Surface expression and ligand-based selectionof cDNAs fused to filamentous phage gene VI. Biotechnology1995; 13:378–382.

83. Crameri R, Suter M. Display of biologically active proteins onthe surface of filamentous phages: a cDNA cloning system forselection of functional gene products linked to the geneticinformation responsible for their production. Gene 1993;137:69–75.

84. Crameri R, Jaussi R, Menz G, Blaser K. Display of expressionproducts of cDNA libraries on phage surfaces. A versatilescreening system for selective isolation of genes by specificgene-product=ligand interaction. Eur J Biochem 1994;226:53–58.

85. Weiss GA, Sidhu SS. Design and evolution of artificial M13coat proteins. J Mol Biol 2000; 300:213–219.

86. Marvin DA, Hale RD, Nave C, Helmer-Citterich M. Molecularmodels and structural comparisons of native and mutantclass I filamentous bacteriophages Ff (fd, f1, M13), If1 andIKe. J Mol Biol 1994; 235:260–286.

87. Kishchenko G, Batliwala H, Makowski L. Structure of a for-eign peptide displayed on the surface of bacteriophage M13.J Mol Biol 1994; 241:208–213.

88. Glucksman MJ, Bhattacharjee S, Makowski L. Three-dimensional structure of a cloning vector. X-ray diffractionstudies of filamentous bacteriophage M13 at 7 A resolution.J Mol Biol 1992; 226:455–470.

89. Hunter GJ, Rowitch DH, Perham RN. Interactions betweenDNA and coat protein in the structure and assembly offilamentous bacteriophage fd. Nature 1987; 327:252–254.



91. Williams KA, Glibowicka M, Li Z, Li H, Khan AR, Chen YM,Wang J, Marvin DA, Deber CM. Packing of coat proteinamphipathic and transmembrane helices in filamentousbacteriophage M13: role of small residues in protein oligomer-ization. J Mol Biol 1995; 252:6–14.

92. Kishchenko GP, Minenkova OO, Ilyichev AA, Gruzdev AD,Petrenko VA. Study of the structure of phage-M13 virionscontaining chimeric B-protein molecules. Mol Biol Engl Trans1991; 25:1171–1176.

93. Weiss GA, Wells JA, Sidhu SS. Mutational analysis of themajor coat protein of M13 identifies residues that controlprotein display. Protein Sci 2000; 9:647–654.

94. Ivanenkov V, Felici F, Menon AG. Targeted delivery of multi-valent phage display vectors into mammalian cells. BiochimBiophys Acta 1999; 1448:463–472.

95. Huie MA, Cheung MC, Muench MO, Becerril B, Kan YW,Marks JD. Antibodies to human fetal erythroid cells from anonimmune phage antibody library. Proc Natl Acad SciUSA 2001; 98:2682–2687.

96. Larocca D, Baird A. Receptor-mediated gene transfer byphage-display vectors: applications in functional genomicsand gene therapy [review]. Drug Discov Today 2001; 6:793–801.

97. Petrenko VA, Vodyanoy VJ. Phage display for detection ofbiological threat agents. J Microbiol Methods 2003; 1768:1–10.

98. Hoogenboom HR, Griffiths AD, Johnson KS, Chiswell DJ,Hudson P, Winter G. Multi-subunit proteins on the surfaceof filamentous phage: methodologies for displaying antibody(Fab) heavy and light chains. Nucleic Acids Res 1991; 19:4133–4137.

99. MacKenzie R, To R. The role of valency in the selection ofanti-carbohydrate single-chain Fvs from phage displaylibraries. J Immunol Methods 1998; 220:39–49.


100. O’Connel D, Becerril B, Roy-Burman A, Daws M, Marks JD.Phage versus phagemid libraries for generation of humanmonoclonal antibodies. J Mol Biol 2002; 321:49–56.

101. de la Cruz VF, Lal AA, McCutchan TF. Immunogenicity andepitope mapping of foreign sequences via genetically engi-neered filamentous phage. J Biol Chem 1988; 263:4318–4322.

102. McConnell SJ, Uveges AJ, Fowlkes DM, Spinella DG. Con-struction and screening of M13 phage libraries displayinglong random peptides. Mol Divers 1996; 1:165–176.

103. Fowlkes DM, Adams MD, Fowler VA, Kay BK. Multipurposevectors for peptide expression on the M13 viral surface.Biotechniques 1992; 13:422–428.

104. Burritt JB, Quinn MT, Jutila MA, Bond CW, Jesaitis AJ.Topological mapping of neutrophil cytochrome b epitopeswith phage-display libraries. J Biol Chem 1995; 270:16974–16980.

105. Burritt JB, Bond CW, Doss KW, Jesaitis AJ. Filamentousphage display of oligopeptide libraries. Anal Biochem 1996;238:1–13.

106. Devlin JJ, Panganiban LC, Devlin PE. Random peptidelibraries: a source of specific protein binding molecules.Science 1990; 249:404–406.

107. O’Neil KT, Hoess RH, Jackson SA, Ramachandran NS,Mousa SA, DeGrado WF. Identification of novel peptideantagonists for GPIIb=IIIa from a conformationally con-strained phage peptide library. Proteins 1992; 14:509–515.

108. Hammer J, Takacs B, Sinigaglia F. Identification of a motiffor HLA-DR1 binding peptides using M13 display libraries.J Exp Med 1992; 176:1007–1013.

109. McLafferty MA, Kent RB, Ladner RC, Markland W. M13 bac-teriophage displaying disulfide-constrained microproteins[review]. Gene 1993; 128:29–36.

110. Malik P, Perham RN. New vectors for peptide display on thesurface of filamentous bacteriophage. Gene 1996; 171:49–51.


111. Malik P, Perham RN. Simultaneous display of different pep-tides on the surface of filamentous bacteriophage. NucleicAcids Res 1997; 25:915–916.

112. Huse WD. Combinatorial antibody expression libraries infilamentous phage. In: Borrebaeck CAK, ed. Antibody Engi-neering: A Practical Guide. New York: W.H. Freeman andCompany, 1991:103–120.

113. Huse WD, Stinchcombe TJ, Glaser SM, Starr L, MacLean M,Hellstrom KE, Hellstrom I, Yelton DE. Application of afilamentous phage pVIII fusion protein system suitable forefficient production, screening, and mutagenesis of F(ab) anti-body fragments. J Immunol 1992; 149:3914–3920.

114. Hoogenboom HR, de Bruine AP, Hufton SE, Hoet RM, ArendsJW, Roovers RC. Antibody phage display technology and itsapplications. Immunotechnology 1998; 4:1–20.

115. Barbas CF III, Kang AS, Lerner RA, Benkovic SJ. Assemblyof combinatorial antibody libraries on phage surfaces: thegene III site. Proc Natl Acad Sci USA 1991; 88:7978–7982.

116. Tang Y, Jiang N, Parakh C, Hilvert D. Selection of linkers fora catalytic single-chain antibody using phage display technol-ogy. J Biol Chem 1996; 271:15682–15686.

117. Kang AS, Barbas CF, Janda KD, Benkovic SJ, Lerner RA.Linkage of recognition and replication functions by assem-bling combinatorial antibody Fab libraries along phagesurfaces. Proc Natl Acad Sci USA 1991; 88:4363–4366.

118. Shin YC, Kim YE. T-J Cho. A novel phage display vector foreasy monitoring of expressed proteins. J Biochem Mol Biol2000; 33:242–248.

119. Soderlind E, Simonsson AC, Borrebaeck CA. Phage displaytechnology in antibody engineering: design of phagemid vec-tors and in vitro maturation systems. Immunol Rev 1992;130:109–124.

120. Ridder R, Schmitz R, Legay F, Gram H. Generation of rabbitmonoclonal antibody fragments from a combinatorial phagedisplay library and their production in the yeast Pichiapastoris. Biotechnology 1995; 13:255–260.


121. Paschke M, Zahn G, Warsinke A, Hohne W. New seriesof vectors for phage display and prokaryotic expression ofproteins. Biotechniques 2001; 30.

122. Zahn G, Skerra A, Hohne W. Investigation of a tetracycline-regulated phage display system. Protein Eng 1999; 12:1031–1034.

123. Bass S, Greene R, Wells JA. Hormone phage: an enrichmentmethod for variant proteins with altered binding properties.Proteins 1990; 8:309–314.

124. Breitling F, Dubel S, Seehaus T, Klewinghaus I, Little M. Asurface expression vector for antibody screening. Gene1991; 104:147–153.

125. Dubel S, Breitling F, Fuchs P, Braunagel M, Klewinghaus I,Little M. A family of vectors for surface display and produc-tion of antibodies. Gene 1993; 128:97–101.

126. Beck E, Zink B. Nucleotide sequence and genome organisa-tion of filamentous bacteriophages fl and fd. Gene 1981; 16:35–58.

127. van Wezenbeek PM, Hulsebos TJ, Schoenmakers JG. Nucleo-tide sequence of the filamentous bacteriophage M13 DNAgenome: comparison with phage fd. Gene 1980; 11:129–148.

128. Baas PD, Jansz HS. Single-stranded DNA phage origins.Curr Top Microbiol Immunol 1988; 136:31–70.

129. Higashitani N, Higashitani A, Horiuchi K. Nucleotidesequence of the primer RNA for DNA replication of filamen-tous bacteriophages. J Virol 1993; 67:2175–2181.

130. Smith GP. Filamentous phage as cloning vectors. In: RodriquezRL, ed. Vectors: A Survey of Molecular Cloning Vectors andTheir Uses. Boston: Butterworth, 1987:61–83.

131. Smith GP, Fernandez AM. Effect of DNA copy number ongenetic stability of phage-displayed peptides. Biotechniques2004; 36(4):610–614, 616, 618.


3

Methods for the Constructionof Phage-Displayed Libraries

FREDERIC A. FELLOUSE

Department of Protein Engineering,Genentech, Inc., South San Francisco,

California, U.S.A.

GABOR PAL

Department of Biochemistry,Eotvos Lorand University,

Budapest, Hungary

I. INTRODUCTION

In this chapter, we present different methods for makinglibraries of recombinant polypeptides. First, we describe howto create highly defined libraries by using synthetic DNA. Sec-ond, we present alternative approaches that can be used tointroduce mutations randomly. In these cases, the positionand nature of the mutations are not defined by the experimen-ter. Both types of methods enable the production of librariesthat can be used to select polypeptides with desired functions.Finally, we describe techniques that have been developed to

111

further increase the diversity of these libraries by eitherrecombination or shuffling.

II. OLIGONUCLEOTIDE-DIRECTEDMUTAGENESIS

Oligonucleotide-directed mutagenesis is a highly controlledmethod for the introduction of mutations into proteins. Thetechnique is usually used when the region to be randomizedis confined to a limited area, but protocols have also beendeveloped to allow for the mutagenesis of entire genes. Byintroducing degeneracy into a synthetic oligonucleotidesequence, the position and degree of randomization can beprecisely controlled. This section presents various strategiesfor the design and synthesis of degenerate oligonucleotidesand methods for the introduction of oligonucleotide librariesinto phage display vectors.

II.A. Strategies for the Design and Synthesisof Degenerate Codons

II.A.1. Saturation Mutagenesis

‘‘Saturation mutagenesis’’ or ‘‘hard randomization’’ refers tothe replacement of one or more codons by a codon thatencodes for all 20 natural amino acids. One degenerate codonthat encodes for all 20 amino acids contains an equimolar mixof all four nucleotides at each position within the codon. TheNNN codon (Table 1) contains all 64 natural codons and thishigh redundancy leads to a highly biased protein diversity,since some amino acids are represented by only one codonwhile others are represented by as many as six codons.Furthermore, three of the unique codons are stop codons,which cause premature polypeptide termination and the lossof functional library members. As a less redundant alterna-tive, the third codon position is allowed to vary either asG=C or G=T (NNK or NNS codons, Table 1), and this resultsin 32 unique codons that cover all 20 amino acids and containonly one stop codon.

112 Fellouse and Pal

Saturation mutagenesis was successfully applied to gen-erate the first synthetic antibody library (1), and it has beenroutinely used for the construction of highly diverse polypep-tide libraries. This simple strategy is very successful in appli-cations where a thorough search of a localized region isrequired, as for example, for antibody hypervariable loopsthat have evolved to display diverse sequences (1) or for naıvepeptide libraries (2–5). The theoretical diversities of librariesincrease exponentially with the number of residues that arerandomized, while the practical diversities accessible through

Table 1 Useful Degenerate Codons

Codon Description Amino acidsa Stop codonsNo. ofcodonsb

NNN All 20 aminoacids

All 20 TAA, TAG,TGA

64

NNK orNNS

All 20 aminoacids

All 20 TAG 32

NNC 15 amino acids A, C, D, F, G,H, I, L, N, P, R,S, T, V, Y

None 16

NWW Charged,hydrophobic

D, E, F, H, I, K,L, N, Q, V, Y

TAA 16

RVK Charged,hydrophilic

A, D, E, G, H, K,N, R, S, T

None 12

DVT Hydrophilic A, C, D, G, N,S, T, Y

None 9

NVT Charged,hydrophilic

C, D, G, H, N, P,R, S, T, Y

None 12

NNT Mixed A, D, G, H, I, L,N, P, R, S, T, V

None 16

VVC Hydrophilic A, D, G, H, N, P,R, S, T

None 9

NTT Hydrophobic F, I, L, V None 4RST Small side chains A, G, S, T None 4TDK Hydrophobic C, F, L, W, Y TAG 6

aIUB code nomenclature (N¼G=A=T=C, K¼G=T, S¼G=C, W¼A=T, R¼A=G,V¼G=A=C, D¼G=A=T).

bThe number of unique codons contained in the degenerate codon. Due to the redun-dancy of the genetic code, the number of unique codons can be higher than thenumber of unique amino acids encoded by a degenerate codon.

Methods for the Construction of Phage-Displayed Libraries 113

phage display (�1011) are limited by bacterial transformationefficiencies (6). Consequently, only six or seven positions canbe subjected simultaneously to saturation mutagenesis witha reasonable probability of completely representing all thepossible amino acid combinations (7) (Table 2). As alterna-tives to saturation mutagenesis, many different strategieshave been developed for the construction of libraries withreduced or more controlled chemical diversity.

II.A.2. Mutagenesis with Spiked Oligonucleotides

It is possible to synthesize mutagenic oligonucleotides bydeliberately contaminating the wild type sequence by adefined level of the other nucleotides. The result is a ‘‘spiked’’oligonucleotide that encodes predominantly a wild type pro-tein sequence but also encodes mutations at a frequencydependent upon the levels of non-wild-type nucleotide con-tamination (8,9). In these synthesis schemes, codons arealtered predominantly by the mutation of only one nucleotide,and consequently, substitutions requiring more than a singlenucleotide change will have a lower probability of occurrence.Libraries produced in this way are highly biased towards thewild type protein sequence, and thus, this strategy is some-times referred to as a ‘‘soft’’ randomization approach since it

Table 2 Diversities of DNA Encoded Protein Libaries

Randomizedpositions

DNAdiversitya(32n)

Proteindiversity(20n)

Requiredlibrary sizeb

1 32 20 1.5� l02

2 1.0� l03 4.0� l02 4.8� l03

3 3.3� l04 8.0� l03 1.5� l05

4 1.1� l06 1.6� l05 4.9� l06

5 3.4� l07 3.2� l06 1.6� l08

6 1.1� l09 6.4� l07 5.0� l09

7 3.4� l010 1.3� l09 1.6� l011

8 1.1� l012 2.6� l010 5.1� l012

aCalculations based on the use of an NNK or NNS degenerate codon.bLibrary size required for a complete representation of all possible amino acidsequences with a 99% confidence using a Poisson distribution.


produces subtle changes in sequence that are particularlyuseful for affinity maturation applications (6).

II.A.3. Mutagenesis with TailoredOligonucleotides

Another strategy for reducing the library diversity is to ran-domize with degenerate codons that encode for only a selectedsubset of the natural amino acids. In this approach, the aminoacid subset is chosen to favor functional polypeptides and isselected on the basis of several possible criteria.

By comparing the sequences of homologous proteins fromdifferent species, one can define a list of amino acid residuesthat are likely to be well tolerated at a particular position.If the three-dimensional structure of the protein is known,rational design methods can be applied to generate a list ofmutations likely to be beneficial. Similarly, certain aminoacids can also be excluded from positions where they arelikely to be deleterious.

Tailored diversity can also be designed based on theresults of previous selections. For example, subsets of posi-tions within a polypeptide can be subjected to saturationmutagenesis in several separate libraries. Statistical analysisof the sequences of functional clones selected from theselibraries can identify positions that are biased towards cer-tain subsets of amino acids (5). This information can thenbe used to design tailored degenerate codons that encode forthe preferred amino acids, and these codons can be used insecond generation libraries.

Many amino acid subsets can be readily designed by sim-ple inspection of the genetic code (Fig. 1) and the use of stoi-chiometric mixtures of nucleotides in degenerate codons(Table 1). However, degenerate codons for some amino aciddistributions cannot be elucidated in this manner. Thus,nonstoichiometric mixtures of bases can be useful to obtainparticular amino acid combinations, and computer algorithmshave been developed for this purpose. Arkin and Youvan (10)analyzed the possible amino acid distributions encoded by allthe possible degenerate codon sets that can be obtained with a


fractional resolution of 10% (i.e., setting the proportion ofeach nucleotide type from 0% to 100% in 10% increments).LaBean and Kauffman (11) developed a semiautomatedapproach to design codon sets that mimic the overall aminoacid compositions characteristic for natural globular proteins.

II.A.4. Mutagenesis with AdvancedOligonucleotide Synthesis Methods

Automated DNA synthesis is traditionally performed on resinbeads packed into a reaction column. After any step of thesynthesis, the beads can be split and repacked into two or sev-eral new columns. Each of these columns can be individually

Figure 1 The genetic code. The 64 genetic codons encode for 20natural amino acids shown in the three letter code. Three codons(TAA, TAG, TGA) encode for stop codons shown as asterisks.


coupled with single or mixed nucleotides and then pooled andrepacked into different columns again (12). In principal, thisresin-splitting method can be used to obtain exact predeter-mined combinations of codons. In contrast, by using standardmethods, synthesis of some codon subsets requires the inclu-sion of codons for additional undesired amino acids. For exam-ple, Fig. 2 illustrates how resin-splitting can be used to obtainonly tyrosine and tryptophan codons. Unfortunately, themethod becomes increasingly complicated as more complexcodon sets are required, and a method that can generate

Figure 2 Oligonucleotide synthesis with (A) a single column or(B) by the split-pool method. Each growing oligonucleotide is cova-lently linked to a bead surface. In the example, the goal is to synthe-size a degenerate codon that encodes for only Trp and Phe. Withsingle column synthesis, codons for two additional amino acids(Cys and Leu) are generated, but split-pool synthesis can be usedto obtain the binomial distribution. Synthetic DNA is typicallysynthesized in the 30–50 direction, but 50–30 synthesis is depictedfor clarity.


any composition of the 20 amino acids requires the use of 10synthesis columns and a tediously frequent resin-splittingand mixing procedure (13,14).

Since amino acids are coded by nucleotide triplets, theideal way to synthesize degenerate oligonucleotides wouldbe the use of trinucleotides as building blocks instead of thetraditional mononucleotides. This would eliminate the codonbiases introduced by the redundant nature of the genetic codeand would allow for the introduction of precise compositionsof the amino acids at positions to be mutagenized. The applic-ability of this technique relies on two prerequisites: availabil-ity of high quality trinucleotide precursors in large quantities,and high efficiency of coupling with the trinucleotide blocks.Several research groups have reported the successful synthe-size and utilization of trinucleotides (5,15–17), and a commer-cial source has recently become available (Glen Research,Sterling, VA). Thus, it seems likely that this method shouldbecome more generally applicable in the near future.

II.B. Integration of Mutagenic Oligonucleotidesinto Phage Display Vectors

To produce libraries from synthetic DNA, it is necessary toinsert the DNA into a suitable phage display vector and tointroduce the resulting recombinant DNA into a bacterialhost for phage production. There are several strategieswhereby this can be achieved. Perhaps the simplest methodinvolves the annealing of a mutagenic oligonucleotide to asingle-stranded DNA (ssDNA) vector and subsequentenzyme-mediated incorporation into a double-stranded DNA(dsDNA) form. Alternatively, dsDNA cassette mutagenesiscan be used and many polymerase chain reaction (PCR) meth-ods have also been developed. The principles behind thesevarious approaches are described below.

II.B.1. The ssDNA Method

The ssDNAmethod is ideally suited to the construction of M13phage libraries, since the viral DNA is packaged in an ssDNAform and can be easily purified from phage particles (6). An


appropriately designed oligonucleotide is annealed to thessDNA template and enzymatically extended and ligated toproduce a dsDNA heteroduplex (Fig. 3). For library construc-tion, the mutagenic oligonucleotide contains degenerateDNA in the region to be mutagenized flanked at both endsby approximately 18 bases that are complementary to theregions preceding and following the target region. Thus,the complementary regions ensure correct annealing aroundthe target region and the degenerate DNA incorporates thedesigned mutations. As oligonucleotides containing up to 100bases can be synthesized with standard methods, a single oli-gonucleotide can introduce mutations in up to approximately20 consecutive codons. Furthermore, multiple oligonucleotidescan be simultaneously annealed to allow for mutations inregions that are far apart in the primary sequence (18,19).

Figure 3 Oligonucleotide-directed mutagenesis with an ssDNAtemplate. (A) A synthetic oligonucleotide (solid line) is annealed tothe ssDNA template (dashed line). The oligonucleotide is designedto encode mutations (shown as asterisks) in the mismatched vari-able region which is flanked by perfectly complementary sequences.(B) Heteroduplex, covalently closed circular dsDNA (CCC-dsDNA)is enzymatically synthesized by T7 DNA polymerase and T4 DNAligase. (C) Heteroduplex dsDNA is introduced into an E. coli hostwhere the mismatched region is repaired to either the wild typeor mutant sequence.


The original ssDNAmutagenesis method (20,21) sufferedfrom low mutation frequencies because the Escherichia colihost preferentially replicated the parental DNA strand ratherthan the in vitro synthesized mutagenic strand. A majoradvance in the method was achieved by producing templatessDNA from E. coli dut–=vng– strains, which contain muta-tions that result in DNA containing significant amounts ofuracil bases in place of thymine bases (22,23). The use of tem-plates of this type results in heteroduplexes in which the wildtype strand contains uracil while the mutagenic strand doesnot. Introduction of these heteroduplexes into E. colidutþ=ungþ results in the preferential inactivation of theuracil-containing strand, and thus, greatly increases themutation frequency. Using optimized protocols based on thisstrategy, high mutation frequencies (>80%) and large librarydiversities (>1010) can be readily achieved (6).

II.B.2. Cassette Methods

A mutagenic cassette is a double-stranded DNA fragment inwhich the region containing mutations is flanked by constantregions that contain restriction enzyme cleavage sites for clon-ing into phage display vectors. The cassette can be synthesizedby different methods, but one of the two strands is alwayssynthesized chemically. The other strand can be synthesizedchemically (24) or by enzyme-mediated extension (25–27). Inthe case of complete chemical synthesis, both strands containdegenerate positions and perfectly complementary annealingoccurs only in the flanking regions. In the case of enzyme-mediated extension, the two strands are complementary sincethe synthetic strand is used as a template for DNA polymer-ase. Most cassette mutagenesis strategies require cloning intoenzymatically cleaved dsDNA vectors, but a method thatachieves cassette cloning with ssDNA vectors also has beendescribed (28).

Cassette mutagenesis methods can achieve mutation effi-ciencies close to 100% provided that the enzymatic cleavagereactions are highly efficient. However, there are inherentlimitations accompanied with these methods. The most


important one is the strict requirement for unique restrictionendonuclease cleavage sites flanking the region to be mutated.Furthermore, while cassette methods are well suited for muta-tions within a short continuous stretch of about 30 codons orless, simultaneous mutagenesis of positions located atgreater distances within a gene would require multiple cas-settes, each with its own requirements for unique flankingrestriction sites. Thus, these methods are best suited formutations in a localized region, which can be obtained witha single cassette.

II.B.3. PCR Methods

Several methods use the PCR as a means for incorporating amutagenic oligonucleotide into a cassette that can then becloned into a phage display vector. In the overlap-extensionmethod, two independent PCRs are performed to amplify twoDNA fragments with overlapping termini (29,30). Mutationsare inserted in both PCR products at the overlapping regionsby using mutagenic primers that also contain a region of com-plementarity (Fig. 4). In a second step, the two PCR productsare annealed to each other and the resulting overlap productis extended to produce a mutagenic cassette that can be clonedinto a vector using restriction sites introduced by the flanking,nonmutagenic primers. A significant advantage of thismethod relative to standard cassette mutagenesis is that therestriction sites need not be proximal to the sites of mutation,since they are contained within the flanking primers.

An alternative ‘‘megaprimer’’ method (31,32) uses threeprimers and two successive rounds of PCR (Fig. 5). In the firstround, the central mutagenic primer and one of the flankingprimers are used to amplify a so-called megaprimer. Themegaprimer is then used in a second PCR with the secondflanking primer to generate the complete mutagenic cassettewith mutations incorporated at the annealing site of the cen-tral primer. Again, the flanking primers are designed to incor-porate restriction sites for convenient cassette cloning.

In the inverse PCR method (33), the primers aredesigned to anneal to the opposite template strands with their


Figure 4 Overlap-extension PCR. Two separate PCRs are per-formed (PCR1 and PCR2). The oligonucleotides encoding the muta-tions (triangles) are partially complementary. The two PCRproducts are mixed, denatured, annealed and extended with DNApolymerase to generate the full-length sequence.

Figure 5 The megaprimer PCR method. A PCR (PCR1) is per-formed with a flanking primer and a central primer encoding formutations (triangle). The PCR product is used as a megaprimerthat anneals to the template and reconstitutes the full-length pro-duct. A second PCR (PCR2) preferentially amplifies the mutatedDNA by using a primer that anneals to a flanking sequence (whitebox) added by the flanking primer used in PCR1.


50 ends near to each other (Fig. 6). One of the primers can bedesigned to incorporate mutations while the other can be per-fectly complementary. A PCR with primers of this type resultsin amplification of the entire vector. Circularization of thisproduct by ligation regenerates the vector and incorporatesthe designed mutations.

III. RANDOM MUTAGENESIS

The oligonucleotide-directed methods target distinct areas ofa gene and are thus ideal for highly controlled mutagenesis.However, methods also have been developed for randommutagenesis throughout a gene, and these can be useful whenit is not clear which region of a protein should be mutated toachieve a particular effect. This section describes methods ofthis type.

Figure 6 Inverse PCR. Primers are designed to anneal to oppositestrands with the two annealing sites close to each other. Polymerasechain reaction results in amplification of the entire vector which canbe regenerated by ligation. Mutations can be incorporated bymismatches in one of the primers, as shown.


III.A. In Vitro Chemical Mutagenesis

Chemical mutagenesis methods were instrumental toolsfor classical genetics experiments that required high muta-tion rates (34). The development of powerful site-directedmethods described above have limited the use of theseless controlled methods in phage display. Nonetheless, chemi-cal mutagenesis may be a viable alternative in specializedapplications.

Many chemicals used for whole organism mutagenesiscan also be used in vitro to modify DNA in a manner thatresults in mutations in vivo. Nitrous acid, hydroxylamine,methoxyamine, sodium bisulfite, and many other chemicalscan be used to induce mutations (35–38). However, one issuewith in vitro chemical mutagenesis is how to confine themutations to the gene of interest so as not to complicatethe later selection cycles with mutations in other regions ofthe vector. For mutagens that act only on ssDNA, a targetsegment can be generated in a dsDNA vector by the succes-sive action of an endonuclease and an exonuclease (39). Alter-natively, following treatment with mutagen, the DNA ofinterest can be excised from the vector and cloned into non-mutagenized vector DNA.

In any case, the method needs to be carefully optimizedin order to achieve the desired levels of mutation. Further-more, because different chemicals produce different types ofmutations, mixtures of chemicals might be required toachieve a sufficiently random distribution of mutations.

III.B. E. coli Mutator Strains

E. coli mutator strains are deficient in DNA editing orproofreading enzymes. The most common mutated genes aremutD, mutS or mutT. The mutD and mutT mutations inducetransitions and transversions, respectively, while the mutSmutation induces both transitions and transversions (40).Mutator strains may contain only a single mutant gene,but strains that contain multiple mutant genes are alsoavailable (41).


The use of mutator strains presents the advantage ofbeing extremely simple; phage need simply be passed througha mutator strain without any chemical or enzymatic manipu-lation. However, it is important to be aware of the instability ofthe mutator strain, as loss of the mutator phenotype presentsa growth advantage for the bacteria (41). In order to confinemutations to the gene of interest, a two-plasmid system hasbeen developed (42); one plasmid expresses an error-proneDNA polymerase I, and the other plasmid bears the gene ofinterest downstream of an origin of replication specificallyrecognized by DNA polymerase I.

III.C. Error-Prone PCR

The thermostableTaqDNApolymerase lacks a 30–50 proofread-ing function, and thus, its error frequency (�10�4 per base) isfar higher than that of in vivoDNA synthesis (�10�9 per base)(43). The error frequency can be further enhanced by usingsuboptimal reaction conditions, and this low fidelity can beexploited to produce fairly random mutations through error-prone PCR.

There are several parameters that can be manipulated toalter the fidelity of Taq DNA polymerase. The buffer canbe altered by changing the pH, changing the ratios of thefour nucleotides, increasing the concentration of Mg2þ ionsor by adding Mn2þ ions (44–46). In addition, the reaction cycleparameters can be adjusted to facilitate errors; the number ofcycles or the extension times can be increased, or the anneal-ing temperature can be lowered (47,48).

Ideally, error-prone PCR would result in perfectly ran-dom substitutions evenly distributed along the entire DNAfragment, but in reality, the process is biased. In general, tran-sitions are more frequent than transversions, and mutationstend to cluster in particular regions of the DNA sequence(49). Another important limitation is that the relatively lowmutation rate results in a strong bias towards amino acid sub-stitutions that arise from single nucleotide mutations. Despitethese drawbacks, error-prone PCR has been used quite exten-sively in in vitro evolution experiments (48,50,51).


IV. COMBINATORIAL INFECTION ANDRECOMBINATION

In vivo recombination strategies have been developed toexpand the diversities of phage display libraries beyond thelimits imposed by E. coli transformation efficiencies. Themethods are particularly effective for heterodimeric proteinssuch as antigen-binding fragments (Fabs). However, varia-tions of the methods also allow for applications with single-chain proteins in which two distinct regions of the sequenceare mutated.

In the simplest recombination method (Fig. 7), twolibraries are created independently and each library codesfor one chain of a heterodimeric protein. One library is car-ried by a phagemid packaged into phage particles, whilethe other is maintained as a plasmid in bacteria. Bacteriacarrying the plasmid library are infected with the phagemidlibrary, and site-specific recombination between the phage-mid and the plasmid forms a new phagemid that containsboth genes of the heterodimeric protein. Thus, a huge librarycan be produced by combining two moderate size libraries,and the size of the library is only limited by the number ofinfected cells.

The concept was first demonstrated (52) by using combi-natorial infection and the site-specific Cre-lox recombinationsystem (53) to generate a Fab that contained light and heavychains encoded by different vectors. The loxP recombinationsites persist after the recombination event, and thus,heterodimeric Fab proteins are ideally suited for this method,because the recombination sites can be introduced into anontranscribed DNA sequence anywhere between the twogenes. Mixed pairs of nonhomologous loxP sequences wereused, as Cre recombinase does not catalyze recombinationbetween the wild type and the mutant loxP sites (54). Eachvector contained one wild type and one mutant sequence toallow recombination between the two vectors while avoidingdeletion of DNA segments through cis recombination. Thesame group subsequently used the strategy to generate aFab library containing 6.5� l010 members (55).


The loxP site is 34 base pairs long, and if it is introducedinto a gene encoding for a monomeric protein, its sequencehas to be in the same reading frame as that of the gene andits translation must not interfere with the function of the pro-tein. In an application of this type, the loxP site was used as alinker between the VL and VH domains of a single-chain vari-able fragment (scFv) (56). In another application, the second

Figure 7 Generation of a Fab library by combinatorial infectionand in vivo recombination (55). Bacteria harboring a plasmid carry-ing a heavy chain library are infected with a phagemid library oflight chains with a constant heavy chain. Subsequent infection witha P1 bacteriophage that provides Cre recombinase results in theexchange of DNA fragments between plasmids and phagemids(represented by scissors). The in vivo recombination produces aphagemid library with 6.5� l010 members with combined lightchain and heavy chain diversity.


complementarity determining region (CDR2) was replaced bya loxP site and separate CDR1 andCDR3 libraries were recom-bined (57). The number of amino acids inserted by the loxP sitehas been reduced by placing the site within a self-splicingintron; following recombination, self-splicing of the intronresulted in a remaining insertion of only 15 base pairs (58).

The in vivo recombination system has also been simpli-fied so as to require only a single vector (59). This strategy

Figure 8 Generation of an scFv library by multiple infection andin vivo recombination (59). An scFv phagemid library with diversityin the light and heavy chains was used to infect E. coli at a veryhigh multiplicity of infection. As a result, each bacterial cell har-bored multiple phagemids and in vivo recombination (representedby scissors) mediated by Cre recombinase resulted in the shufflingof light and heavy chains from different phagemids. The resultingrecombined library had an estimated diversity of 3� l011 members.


was applied to an scFv library; mutant and wild type loxPsites flanked the region encoding the VH domain and the coatprotein fusion (Fig. 8). Bacteria expressing Cre recombinasewere infected with extremely high concentrations of phage,and this resulted in individual cells being infected by multiplephage particles. Subsequent recombination resulted in VH

domain exchange between different vectors, and a recom-bined library of 3� l011 members was obtained from aprimary library of 7� l07 members.

V. DNA SHUFFLING

Oligonucleotide-directed and random mutagenesis methodscan be used to generate mutations but they lack an importantelement of natural evolution, namely they do not allow forrecombination of mutations from different selected sequences.Genetic recombination is a powerful means whereby benefi-cial traits engendered by individual mutations can be com-bined to potentially generate even greater beneficial effects.DNA shuffling techniques have been developed to mimicgenetic recombination, and these methods have greatlyenhanced the power of in vitro evolution strategies.

In the original DNA shuffling method (60,61) (Fig. 9A), aDNA fragment containing the gene of interest was digestedwith Dnase I to yield a pool of small, random fragments.These fragments were then reassembled with a self-primingPCR in which the fragments primed each other to regeneratethe full-length gene. By using a pool of DNA fragments thatcontained mutations selected for a particular trait, the DNAshuffling procedure resulted in a new pool that containednot only the starting sequences but also products thatresulted from the recombination of different sequences withinthe pool. In this way, mutations from different clones wereeffectively combined to generate genes containing multiplemutations that could then be tested for further improvementsin the selected trait. Additional mutations could also be intro-duced by the error-prone nature of DNA replication by TaqDNA polymerase. DNA shuffling has also been applied to


Figure 9 Principles and limitations of DNA shuffling. (A) Parental genes are fragmented. The fragmentsare mixed and reassembled by selfpriming PCR which results in the shuffling of sequences from differentparents. (B) Mutations that are close together will only recombine rarely (low resolution problem) and willtend to remain clustered together (dashed box). (C) If the homology between parental genes is too low, thefragments will reassemble mainly as the original parental genes.

130

Fellouse

andPal

combine sequences of related homologous genes rather thanmutants of a single gene, and this process of ‘‘DNA familyshuffling’’ has been very effective in generating novel func-tional proteins that combine sequence elements from closelyrelated natural homologues (62).

Although very straightforward and used with great suc-cess, the original DNA shuffling technique suffers from sev-eral limitations. First of all, the Dnase I cleavage reaction isnot perfectly random, as the enzyme prefers to hydrolyzedsDNA at sites adjacent to pyrimidine nucleotides (63). Thenumber of crossovers that can be created is also restrictedby limitations on the fragment size, which has to providelarge enough overlap to ensure stable annealing betweenthe fragments. Moreover, crossover occurs preferentially atregions of high sequence identity (64,65). As a result of theselimitations, blocks of parental sequences tend to be conservedas the method provides a low crossover resolution (Fig. 9B). Inthe case of DNA family shuffling of natural homologues, thesequence identity must exceed a certain threshold (�70%) toallow for productive annealing between homologues (66).Below the threshold value, very little recombination occursand the original parental genes are the major products ofthe reaction (Fig. 9C). Several alternatives and improvementsto the DNA shuffling method have been developed to addressthese limitations, as described below.

V.A. Methods for Improved CrossoverResolution

Several methods have been developed in an effort to increasethe frequency of crossovers in DNA shuffling. In the random-priming in vitro recombination strategy, random hexanucleo-tides are annealed to two parental templates and used toprime DNA synthesis (67). The short fragments are then reas-sembled, by self-priming PCR, and it was shown that up to sixcrossovers were obtained. Alternatively, a staggered exten-sion process (StEP) utilizes primers that anneal to one endof the parental genes and PCR with very short annealing


and extension steps (Fig. 10) (68). As a result, only shortextensions occur at each cycle and the growing strands canswitch templates multiple times as the replication of thefull-length genes requires multiple cycles. In a method called‘‘random chimeragenesis on transient templates’’ (RACHITT),a uracil-containing parental ssDNA strand is used as a tem-plate onto which fragments generated by Dnase digestion

Figure 10 The staggered extension process (StEP) (68). A primeris annealed to one end of each parental template and is extended invery short cycles of denaturation and reannealing. As a result, theelongating strand pairs with different templates in each extensioncycle and the template switching process results in chimeric genes.


are annealed (Fig. 11) (69). Unhybridized ends on the anne-aled fragments are removed by nuclease digestion, gaps arefilled in by DNA polymerase and full-length genes are regen-erated by sealing nicks with DNA ligase. The transienttemplate is rendered unamplifiable by treatment with ura-cil-DNA glycosylase and PCR can be used to amplify the chi-meric genes. It was demonstrated that this method produced

Figure 11 Random chimeragenesis on transient templates(RACHITT) (69). Fragments from multiple genes are annealed ontoa uracil-containing ssDNA template. Unannealed flaps are removedby nuclease treatment, the gaps are filled and the nicks are ligated.The ssDNA template is inactivated by treatment with uracil-DNAglycosylase and the reassembled chimeric genes are amplifiedby PCR.


much higher crossover frequencies in comparison with con-ventional DNA shuffling.

V.B. Incorporation of Synthetic DNA

The earliest reports of DNA shuffling demonstrated thatdiversity could be increased by the addition of syntheticoligonucleotides that were incorporated into the reassembledchimeric genes (60,70). Indeed, several reports have

Figure 12 Degenerate homoduplex recombination (DHR) (73).Conserved regions are identified by sequence alignment of homolo-gous genes. Synthetic oligonucleotide are synthesized as eitherextendable top-strands or as nonextendable bottom-strands. Thebottom-strand oligonucleotides serve to assemble the top-strand oli-gonucleotides (dashed boxes). Enzyme-mediated extension and liga-tion generates full-length chimeric genes which can be amplified byPCR.


demonstrated that it is feasible to perform DNA shufflingwith synthetic oligonucleotides as the only source of DNA(71,72), as exemplified by the degenerate homoduplex recom-bination (DHR) method (73). In this method, ‘‘top-strand’’ oli-gonucleotides are synthesized so that together they span theentire length of a gene with the exception of small gapsbetween the oligonucleotides (Fig. 12). The gaps are bridgedby shorter ‘‘bottom-strand’’ oligonucleotides that are designedto anneal to the ends of two adjacent top-strand oligonucleo-tides. The oligonucleotides are modified during chemicalsynthesis such that only the top-strand oligonucleotides arecompetent for enzyme-mediated extension and ligation. Chi-meric gene assembly is achieved by annealing all of the oligo-nucleotides together, followed by extension and ligation to fillin the gaps to produce full-length, recombined genes.

Methods of this type combine the precision ofoligonucleotide-directed mutagenesis with the combinatorialpower of DNA shuffling, and they provide significant advan-tages over conventional DNA shuffling. The use of degeneratesynthetic DNA enables precise design of sequence variationsand greatly increases the levels of diversity that can be intro-duced. Secondly, the crossovers are generated in a designedmanner through chemical synthesis, and the process can bemore precisely controlled to achieve optimal, unbiased cross-over resolution.

REFERENCES

1. Barbas CF III, Bain JD, Hoekstra DM, Lerner RA. Semisyn-thetic combinatorial antibody libraries: a chemical solution tothe diversity problem. Proc Natl Acad Sci USA 1992;89:4457–4461.





5. Deshayes K, Schaffer ML, Skelton NJ, Nakamura GR,Kadkhodayan S, Sidhu SS. Rapid identification of small bind-ing motifs with high-throughput phage display: discovery ofpeptidic antagonists of IGF-1 function. Chem Biol 2002; 9:495–505.

6. Sidhu SS, Lowman HB, Cunningham BC, Wells JA. Phage dis-play for selection of novel binding peptides. Methods Enzymol2000; 328:333–363.

7. Clackson T, Wells JA. In vitro selection from protein andpeptide libraries. Trends Biotechnol 1994; 12:173–184.

8. Hutchison CA III, Nordeen SK, Vogt K, Edgell MH. A completelibrary of point substitution mutations in the glucocorticoidresponse element of mouse mammary tumor virus. Proc NatlAcad Sci USA 1986; 83:710–714.

9. Hermes JD, Parekh SM, Blacklow SC, Koster H, Knowles JR.A reliable method for random mutagenesis: the generationof mutant libraries using spiked oligodeoxyribonucleotideprimers. Gene 1989; 84:143–151.

10. Arkin AP, Youvan DC. Optimizing nucleotide mixtures toencode specific subsets of amino acids for semi-random muta-genesis. Biotechnology (NY) 1992; 10:297–300.

11. LaBean TH, Kauffman SA. Design of synthetic gene librariesencoding random sequence proteins with desired ensemblecharacteristics. Protein Sci 1993; 2:1249–1254.

12. Glaser SM, Yelton DE, Huse WD. Antibody engineering bycodon-based mutagenesis in a filamentous phage vector sys-tem. J Immunol 1992; 149:3903–3913.

13. Huse WP, Yelton DE, Glaser SM. Increased antibody affinityand specificity by codon-based mutagenesis. Int Rev Immunol1993; 10:129–137.

14. Neuner P, Cortese R, Monaci P. Codon-based mutagenesisusing dimer-phosphoramidites. Nucleic Acids Res 1998; 26:1223–1227.


15. Ono A, Matsuda A, Zhao J, Santi DV. The synthesis of blockedtriplet-phosphoramidites and their use in mutagenesis.Nucleic Acids Res 1995; 23:4677–4682.

16. Sondek J, Shortle D. A general strategy for random insertionand substitution mutagenesis: substoichiometric coupling oftrinucleotide phosphoramidites. Proc Natl Acad Sci USA1992; 89:3581–3585.

17. Virnekas B, Ge L, Pluckthun A, Schneider KC, Wellnhofer G,Moroney SE. Trinucleotide phosphoramidites: ideal reagentsfor the synthesis of mixed oligonucleotides for random muta-genesis. Nucleic Acids Res 1994; 22:5600–5607.

18. Pal G, Kossiakoff AA, Sidhu SS. The functional binding epi-tope of a high affinity variant of human growth hormonemapped by shotgun alanine-scanning mutagenesis: Insightsinto the mechanisms responsible for improved affinity. JMol Biol 2003; 332:195–204.

19. Vajdos FF, Adams CW, Breece TN, Presta LG, de Vos AM,Sidhu SS. Comprehensive functional maps of the antigen-binding site of an anti-ErbB2 antibody obtained with shotgunscanning mutagenesis. J Mol Biol 2002; 320:415–428.

20. Zoller MJ, Smith M. Oligonucleotide-directed mutagenesis ofDNA fragments cloned into M1 3 vectors. Methods Enzymol1983; 100:468–500.

21. Zoller MJ, Smith M. Oligonucleotide-directed mutagenesisusing M13-derived vectors: an efficient and general procedurefor the production of point mutations in any fragment of DNA.Nucleic Acids Res 1982; 10:6487–6500.

22. Kunkel TA. Rapid and efficient site-specific mutagenesis with-out phenotypic selection. Proc Natl Acad Sci USA 1985; 82:488–492.

23. Kunkel TA, Roberts JD, Zakour RA. Rapid and efficient site-specific mutagenesis without phenotypic selection. MethodsEnzymol 1987; 154:367–382.

24. Wells JA, Vasser M, Powers DB. Cassette mutagenesis: an effi-cient method for generation of multiple mutations at definedsites. Gene 1985; 34:315–323.


25. Oliphant AR, Nussbaum AL, Struhl K. Cloning of random-sequence oligodeoxynucleotides. Gene 1986; 44:177–183.

26. Reidhaar-Olson JF, Bowie JU, Breyer RM, Hu JC, Knight KL,Lim WA, Mossing MC, Parsell DA, Shoemaker KR, Sauer RT.Random mutagenesis of protein sequences using oligonucleo-tide cassettes. Methods Enzymol 1991; 208:564–586.

27. Goldman ER, Youvan DC. An algorithmically optimized com-binatorial library screened by digital imaging spectroscopy.Biotechnology (NY) 1992; 10:1557–1561.

28. Bonnycastle LL, Mehroke JS, Rashed M, Gong X, Scott JK.Probing the basis of antibody reactivity with a panel of con-strained peptide libraries displayed by filamentous phage. JMol Biol 1996; 258:747–762.

29. Higuchi R, Krummel B, Saiki RK. A general method of in vitropreparation and specific mutagenesis of DNA fragments: studyof protein and DNA interactions. Nucleic Acids Res 1988;16:7351–7367.

30. Ho SN, Hunt HD, Horton RM, Pullen JK, Pease LR. Site-directed mutagenesis by overlap extension using the polymer-ase chain reaction. Gene 1989; 77:51–59.

31. Kammann M, Laufs J, Schell J, Gronenborn B. Rapid inser-tional mutagenesis of DNA by polymerase chain reaction(PCR). Nucleic Acids Res 1989; 17:5404.

32. Ke SH, Madison EL. Rapid and efficient site-directed muta-genesis by single-tube ‘megaprimer’ PCR method. NucleicAcids Res 1997; 25:3371–3372.

33. Hemsley A, Arnheim N, Toney MD, Cortopassi G, Galas DJ. Asimple method for site-directed mutagenesis using the poly-merase chain reaction. Nucleic Acids Res 1989; 17:6545–6551.

34. De Lucia P, Cairns J. Isolation of an E. coli strain with amutation affecting DNA polymerase. Nature 1969; 224:1164–1166.

35. Solnick D. An adenovirus mutant defective in splicing RNAfrom early region 1A. Nature 1981; 291:508–510.

36. Chu CT, Parris DS, Dixon RA, Farber FE, Schaffer PA. Hydro-xylamine mutagenesis of HSV DNA and DNA fragments:


introduction of mutations into selected regions of the viralgenome. Virology 1979; 98:168–181.

37. Busby S, Irani M, Crombrugghe B. Isolation of mutantpromoters in the Escherichia coli galactose operon using localmutagenesis on cloned DNA fragments. J Mol Biol 1982; 154:197–209.

38. Kadonaga JT, Knowles JR. A simple and efficient method forchemical mutagenesis of DNA. Nucleic Acids Res 1985; 13:1733–1745.

39. Shortle D, Botstein D. Directed mutagenesis with sodiumbisulfite. Methods Enzymol 1983; 100:457–468.

40. Cox EC. Bacterial mutator genes and the control of sponta-neous mutation. Annu Rev Genet 1976; 10:135–156.

41. Greener A, Callahan M, Jerpseth B. An efficient randommutagenesis technique using an E. coli mutator strain. MolBiotechnol 1997; 7:189–195.

42. Camps M, Naukkarinen J, Johnson BP, Loeb LA. Targetedgene evolution in Escherichia coli using a highly error-proneDNA polymerase I. Proc Natl Acad Sci USA 2003; 100:9727–9732.

43. Zakour RA, Loeb LA. Site-specific mutagenesis by error-directed DNA synthesis. Nature 1982; 295:708–710.

44. Vartanian JP, Henry M, Wain-Hobson S. HypermutagenicPCR involving all four transitions and a sizeable proportionof transversions. Nucleic Acids Res 1996; 24:2627–2631.

45. Zhou YH, Zhang XP, Ebright RH. Random mutagenesis ofgene-sized DNA molecules by use of PCR with Taq DNApolymerase. Nucleic Acids Res 1991; 19:6052.

46. Fromant M, Blanquet S, Plateau P. Direct random mutagen-esis of gene-sized DNA fragments using polymerase chainreaction. Anal Biochem 1995; 224:347–353.

47. Leung DW, Cachianes G, Kuang WJ, Goeddel DV, Ferrara N.Vascular endothelial growth factor is a secreted angiogenicmitogen. Science 1989; 246:1306–1309.

48. Cadwell RC, Joyce GF. Randomization of genes by PCR muta-genesis. PCR Methods Appl 1992; 2:28–33.


49. Botstein D, Shortle D. Strategies and applications of in vitromutagenesis. Science 1985; 229:1193–1201.

50. Moore JC, Arnold FH. Directed evolution of a para-nitrobenzylesterase for aqueous–organic solvents. Nat Biotechnol 1996;14:458–467.

51. Zhao H, Arnold FH. Optimization of DNA shuffling for highfidelity recombination. Nucleic Acids Res 1997; 25:1307–1308.

52. Waterhouse P, Griffiths AD, Johnson KS, Winter G. Combina-torial infection and in vivo recombination: a strategy for mak-ing large phage antibody repertoires. Nucleic Acids Res 1993;21:2265–2266.

53. Sternberg N, Hamilton D. Bacteriophage P1 site-specificrecombination. I. Recombination between loxP sites. J MolBiol 1981; 150:467–486.

54. Hoess RH, Wierzbicki A, Abremski K. The role of the loxPspacer region in P1 site-specific recombination. Nucleic AcidsRes 1986; 14:2287–2300.

55. Griffiths AD, Williams SC, Hartley O, Tomlinson IM,Waterhouse P, Crosby WL, Kontermann RE, Jones PT, LowNM, Allison TJ, et al. Isolation of high affinity human antibo-dies directly from large synthetic repertoires. EMBO J 1994;13:3245–3260.

56. Tsurushita N, Fu H, Warren C. Phage display vectors for invivo recombination of immunoglobulin heavy and light chaingenes to make large combinatorial libraries. Gene 1996; 172:59–63.

57. Davies J, Riechmann L. An antibody VH domain with a lox-Cresite integrated into its coding region: bacterial recombinationwithin a single polypeptide chain. FEBS Lett 1995; 377:92–96.

58. Fisch I, Kontermann RE, Finnern R, Hartley O,Soler-Gonzalez AS, Griffiths AD, Winter G. A strategy ofexon shuffling for making large peptide repertoires displayedon filamentous bacteriophage. Proc Natl Acad Sci USA 1996;93:7761–7766.

59. Sblattero D, Bradbury A. Exploiting recombination in singlebacteria to make large phage antibody libraries. Nat Biotech-nol 2000; 18:75–80.


60. Stemmer WP. DNA shuffling by random fragmentation andreassembly in vitro recombination for molecular evolution.Proc Natl Acad Sci USA 1994; 91:10747–10751.

61. Stemmer WP. Rapid evolution of a protein in vitro by DNAshuffling. Nature 1994; 370:389–391.

62. Crameri A, Raillard SA, Bermudez E, Stemmer WP. DNAshuffling of a family of genes from diverse species acceleratesdirected evolution. Nature 1998; 391:288–291.

63. Murakami H, Hohsaka T, Sisido M. Random insertion anddeletion of arbitrary number of bases for codon-based randommutation of DNAs. Nat Biotechnol 2002; 20:76–81.

64. Moore GL, Maranas CD, Lute S, Benkovic SJ. Predictingcrossover generation in DNA shuffling. Proc Natl Acad SciUSA 2001; 98:3226–3231.

65. PatrickWM,FirthAE,Blackburn JM.User-friendly algorithmsfor estimating completeness and diversity in randomizedprotein-encoding libraries. Protein Eng 2003; 16:451–457.

66. Joern JM, Meinhold P, Arnold FH. Analysis of shuffled genelibraries. J Mol Biol 2002; 316:643–656.

67. Shao Z, Zhao H, Giver L. Arnold FH. Random-priming in vitrorecombination: an effective tool for directed evolution. NucleicAcids Res 1998; 26:681–683.

68. Zhao H, Giver L, Shao Z, Affholter JA, Arnold FH. Molecularevolution by staggered extension process (StEP) in vitrorecombination. Nat Biotechnol 1998; 16:258–261.

69. Coco WM, Levinson WE, Crist MJ, Hektor HJ, Darzins A,Pienkos PT, Squires CH, Monticello DJ. DNA shufflingmethod for generating highly recombined genes and evolvedenzymes. Nat Biotechnol 2001; 19:354–359.

70. Crameri A, Stemmer WP. Combinatorial multiple cassettemutagenesis creates all the permutations of mutant andwild-type sequences. Biotechniques 1995; 18:194–196.

71. Ness JE, Kim S, Gottman A, Pak R, Krebber A, Borchert TV,Govindarajan S, Mundorff EC, Minshull J. Synthetic shufflingexpands functional protein diversity by allowing amino acids torecombine independently. Nat Biotechnol 2002; 20:1251–1255.


72. Zha D, Eipper A, Reetz MT. Assembly of designed oligonucleo-tides as an efficient method for gene recombination: a new toolin directed evolution. Chembiochem 2003; 4:34–39.

73. Coco WM, Encell LP, Levinson WE, Crist MJ, Loomis AK,Licato LL, Arensdorf JJ, Sica N, Pienkos PT, Monticello DJ.Growth factor engineering by degenerate homoduplex genefamily recombination. Nat Biotechnol 2002; 20:1246–1250.


4

Selection and Screening Strategies

MARK S. DENNIS

Department of Protein Engineering, Genentech, Inc.,South San Francisco, California, U.S.A.

I. INTRODUCTION

A multitude of selection and screening strategies for phage-displayed libraries have been developed since the techniquewas first described by Scott and Smith (1). Each relies on theability to discriminate desired phage from the library eitherthrough selective capture to allow the removal of unboundphage or through selective elution, as in the case of substratephage to specifically elute desired phage (see Chapter 7). Ulti-mately, the degree to which target directed phage can be sepa-rated from the rest of the phage library will determine theeffectiveness of the panningmethod. This chapter summarizesmany of these strategies and discusses the considerations thatmust be kept in mind when using them.

143

II. GENERAL CONSIDERATIONS

The selection of phage-displayed ligands to a specific targetrequires that the target be presented in a native form andat a sufficient concentration to allow enrichment over back-ground binding phage. If the target concentration is too low,the number of selectively bound phage recovered following around of selection will have no amplification advantage rela-tive to the background binding phage and will thus nevertakeover the library population. Particularly in the initialrounds of selection, the concentration of any given phage inthe library is extremely low and thus the rate of phage bind-ing to target is relatively slow. Capture can be improved byexposure of the phage pool to the target for several hoursat room temperature or overnight at 4�C.

The anticipated affinity of the interaction between thedisplayed ligand and the target is also an important consid-eration (Table 1). If the interaction between a ligand and itstarget is very tight, a target presented at a relatively highconcentration or the displayed ligand presented in a poly-valent format on phage will make selection of tighter bindingvariants more difficult. Inversely, if the interaction is likely tobe weak, as with naıve peptide libraries, polyvalent display ofthe ligand or the presentation of a relatively high concen-tration of target is desired in order to boost the chances offinding an interaction.

Another important aspect of sorting is the number ofphage to include in the selection round. This will depend uponthe diversity and valence of the library. Generally, when

Table 1 Suggested Phage Valence and Target Concentration forSelections

Valence Target concentration (nM) Affinities selected

Polyvalent 1–10 0.1–100 mMMonovalent 100 0.1–10mMMonovalent 1–10 0.1–100nMMonovalent 0.1 10–100pM

144 Dennis

sorting, enough phage should be added so that every memberof the library is represented 100–1000 times. Monovalentlydisplayed libraries require higher representation dependingupon the ligand display level relative to polyvalent phagelibraries for which every phage particle is expected to presentat least a few copies of the displayed ligand. Bass et al. (2), forexample, have estimated that monovalently displayed growthhormone is present on about 10% of the phage, thus a 100-foldover representation of each library member results in eachbeing present only 10 times.

To reduce nonspecific phage binding, irrelevant proteins,such as bovine serum albumin (BSA), gelatin, casein, ovalbu-min or powdered milk are often used to coat the surfaces inthe sorting reaction and can also be included in the sorting buf-fer along with nonionic detergents such as Tween 20 or TritonX-100. Occasionally, phage libraries containmembers thatwillbind to these blocking agents, and this results in a high back-ground after a few rounds of selection and amplification. Insuch cases, a change or rotation of blocking agents and repeti-tion of the previous round of selection generally avoids orreduces this increase in background.Obviously,when choosinga sorting buffer, the stability of the target must be consideredas well as any requirements for reducing agent, divalentcations, or other cofactors. Additionally, trace amounts ofbiotin present in casein or powdered milk can interfere withselections using an avidin capture (see below). In selectionstrategies discussed below, the presentation of the target in adiverse environment can serve to weed out nonspecific phage.

Depending upon the method employed, the selectionstringency is an important consideration. This is particularlytrue in the first rounds of selection when the library diversityis the greatest and the representation of each member is thelowest. In order to maintain this diversity through successiverounds, stringency should begin low and gradually increasewith successive rounds as diversity is reduced and the popula-tion of selectively bound phage increases. Thus in the earlyrounds, efforts to maximize the capture of all potentiallyinteresting clones should be made. As enrichment of selectivebinding clones is observed in later rounds, stringency can be

Selection and Screening Strategies 145

increased. An effective means of increased selection strin-gency is through reduction of the target concentration, caus-ing only the tightest binding clones to be selected. However,a minimum target concentration—one that provides a highernumber of selective versus background phage to be recov-ered—is required for enrichment to occur.

Besides selecting phage based upon binding affinity,phage-displayed libraries can be directed to discriminatebetween two or more closely related targets using a techniquecalled competitive selection. Competitive selection involvesproviding an undesired target in solution while capturing dis-criminating phage that are bound to the desired immobilizedtarget. This technique has been used successfully to increasethe selectivity of a general serine protease inhibitor for the coa-gulation protease, FactorVIIa (FVIIa) (3), and also, to generatereceptor-selective variants of atrial natriuretic peptide andvascular endothelial growth factor (4,5). In many cases, vari-ants exhibiting greater than 1000-fold selectivity have beenobtained. To establish initial selection conditions, the bindingof phage bearing the wild-type ligand to the desired immobi-lized target is monitored in the presence of increasing concen-trations of an undesired competing target. The concentrationof soluble competing target to add during the selection dependsupon its affinity for the initial lead and is determined empiri-cally. The point at which 75–95% of the phage are preventedfrombinding to the immobilized target provides a good startingconcentration of competing target to add in the first round ofselection (see Fig. 1). For successive rounds, an increase inenrichment can be countered with an increase in stringency,generated by increasing the concentration of competing target.In many cases, subtle sequence changes to the binding inter-face can dramatically alter ligand-binding specificity.

III. THE SELECTION PROCESS

III.A. Washing

As mentioned above, in the early rounds of selection, whenlibrary diversity is high and the number of desired clones is

146 Dennis

low, it is important to maximize the recovery of desired clones.Thus, stringent washing to lower background binding phagein the initial rounds of selection is generally not prudent.Wash stringency can be increased in later rounds, as diversityis reduced through the loss of nonbinding clones and the popu-lation of desired clones is increased. This can be achievedthrough multiple washings, increased washing times or byadding increasing concentrations of a competing ligand tothe wash buffer to allow clones with slower dissociation rateconstants to be selected.

Figure 1 Establishing conditions for competitive selection ofphage binding to an immobilized target. Phage displaying thewild-type ligand are added to target (TF �FVIIa) coated wells con-taining increasing concentrations of a competitive target (FXIa) insolution. Selection conditions for libraries based on the displayedligand should begin with the concentration of competitive targetrequired to reduce phage capture by 75–95% (indicated by arrow)and is dependent on the affinity of the ligand for the competitivetarget. As the rounds of selection progress and enrichment isobserved, the concentration of competing target can be increasedto further drive the selection of specific binding variants.


III.B. Elution Methods

Themost commonlyusedmethod for recoveringphagebound toa target is through simple disruption of the binding interactionbetween the displayed ligand and the target by altering the pH.Other disruptive techniques are also suitable as long as theyare reversible to the point that they do not interfere with theamplification of the phage in Escherichia coli (i.e., infection).Although this method generally does not provide complete elu-tion of the bound phage, sufficient phage are eluted to allow foramplification and good enrichments are achieved.

The use of a known ligand to selectively displace boundphage can be successful, although very high concentrationsof ligand and long elution times are generally required. Thisis particularly true for tight binding ligands with slow-offrates. Ligand elution is unlikely to work with ligands dis-played on protein 8 (p8) as a result of the high display andavidity provided by polyvalent display.

Selective proteolytic elution is another useful methodthat has been used to identify selective ligands. For the selec-tion of peptides that bound to the erythropoietin receptor, thetarget was presented as an immunoadhesin (6). This can fre-quently result in unintended selection of clones that bind tothe Fc region of the immunoadhesin. To avoid selection ofthese non-target-directed clones, a thrombin cleavage sitewas introduced between the Fc and the erythropoietin recep-tor. Thus, upon the addition of thrombin and release of theimmobilized target, phage that bound selectively to the recep-tor could be selectively eluted (6).

III.C. Amplification

Following elution, the eluted pool of phage can be amplified inE. coli prior to further rounds of selection. This amplificationprocess can lead to artifacts such as the biased production ofbetter expressing clones or faster growing ‘‘monster’’ phageclones that lack any displayed ligand. Generally, however, ifspecific binding clones are present in the library and are cap-tured in sufficient numbers, these artifacts remain a minorcomponent of the amplified library.

148 Dennis

To avoid a loss of clonal diversity, an excess of rapidlygrowing E. coli host must be provided to capture all the phagepresent in the pool. Following infection with the eluted phage,helper phage is added for the production of new phageparticles.

III.D. Monitoring the Selection Process

By monitoring the enrichment ratio (the ratio of clones elutedfrom a target coated well to clones eluted from a blank well),the progress of the selection process can be tracked. Withnaıve libraries, enrichment is usually not observed duringthe first two rounds of selection; usually in the third roundthe enrichment ratio is greater than 10 and can be as highas 1000 or more in later rounds. For biased libraries, or thosebased on a pre-existing ligand, enrichment may be observedsooner. The number of phage added and recovered during atypical phage selection experiment is shown in Table 2.

III.E. Sequence Analysis

The decision of when to sequence clones from selectedlibraries is based on two general strategies and depends uponthe nature of the library and the information desired. Formaturing existing phage–target associations, such as ligand–receptor interactions, often only the sequence of the best vari-ant is important. Generally, four to six rounds of selectionmay be required to reduce the population to a handful of

Table 2 Phage (cfu) Present in Pools from a Typical PanningExperiment

Phage pool Round 1 Round 2 Round 3 Round 4

Starting phage library diversity 1� 109 5� 105 1� 106 1� 107

Phage input 1� 1011

Target eluted phage 1� 105 5� 105 1� 106 1� 107

Nontarget eluted phage 1� 105 1� 105 1� 105 1� 105

Enrichment 1X 5X 10X 100XAmplified phage pool 1� 1014


‘‘winners.’’ Since phage selection is a function of display level(i.e., expression) as well as binding affinity, excessive roundsof selection can lead to artifacts as mentioned above. Forexample, identical clones or ‘‘siblings’’ present at a high fre-quency may result not only from enhanced binding properties,but also from higher expression and display. Thus, sequencingclones from a selection round that still retains some diversesequences followed by an assessment of the purified ligandsis recommended to avoid missing the best variant. In contrast,when searching for an initial lead, as with naıve libraries,often a diverse set of potential leads can be useful. Here,examination of sequences following just two to three roundsof selection can yield a surprising array of initial leads. In thiscase, a large number of potential leads can be rapidly evalu-ated by assessing specific phage binding en masse beforeprogressing further (7).

While sequence information can be useful and interest-ing, what ultimately matters is a ligand possessing the prop-erties being sought. Several individual phage clones can beselected and tested for target-specific binding. Although diffi-cult to achieve when using polyvalent phage, monovalentphage can be assessed for specific binding and an estimateof affinity can be determined by titrating phage binding withsoluble target (8).

IV. SELECTIONS METHODS

IV.A. Purified Targets

The most frequently used method of screening against puri-fied protein targets involves direct immobilization on a solidsupport such as a bead or microtiter plate. This enables thephysical separation of bound and unbound phage simply bywashing the support. Numerous supports have been used,including modified affinity resins, glass beads, modified mag-netic beads, covalently modified plastics, and chelatingmatrices. Supports should be chosen based upon their lowbackground for nonspecific phage binding and their abilityto present the target in a native conformation and at a

150 Dennis

desirable concentration. Characteristics of the target mole-cule, such as stability, solubility, and availability of freethiols, free amines, or carbohydrates for covalent matrixattachment, are also important considerations. For example,mutations engineered on the surface of a target protein mayprovide a means to chemically modify a specific site to facili-tate capture and present the target in a specific orientation(9). Protein complexes can be screened through direct immo-bilization of the complex or through the specific capture ofthe target through an immobilized associating partner aswas demonstrated in the capture of FVIIa by immobilizedtissue factor (TF) to generate an immobilized TF �FVIIacomplex (10).

Special consideration should be made when panningphage libraries against biotinylated targets captured on avi-din-coated surfaces, since this can lead to the unintendedselection of biotin mimics (11,12). To avoid the selection of bio-tin mimics such as the tripeptide sequence HPQ, excess biotincan be added following capture of the immobilized target toblock remaining free biotin sites prior to the addition of thephage library. In addition, rotation between avidin, streptavi-din, or neutravidin as the capture medium between succes-sive rounds of selection can prevent selection of clonesselective for other regions of these proteins (unpublisheddata).

Alternatively, phage binding to biotinylated targets canbe performed in solution followed by a very brief selective cap-ture of the target on avidin-coated plates or magnetic beads(13). Magnetic beads can be used to reduce avidity effectsresulting from displaying multiple copies of the ligand onphage since independent binding interactions are not tied toa matrix. The brief capture often leads to a dramatically lowerbackground and the ability to bind target in solution providesa convenient way to control the target concentration. Impor-tantly, selection and blocking buffers containing traceamounts of biotin (present in casein and powdered milk)should be avoided.

As evident in Table 1, selections for ligand affinities inthe subnanomolar range are difficult. By utilizing a solution


binding strategy, the target concentration can be reduced toless than 1nM. This in concert with increased stringency,such as a dramatically extended wash period, can lead tothe selection of affinities in the picomole range. When targetconcentrations below 1nM are used, however, very low speci-fic phage titers are recovered. Thus, this method should onlybe used in later selection rounds after diversity has beenreduced and the populations of binding phage are high.Although phage will be recovered, the values determined forenrichment are not likely to be meaningful, and thus, furthercharacterization of selected clones is required to identify thosewith desired properties.

IV.B. Cell-Surface Targets

Several methods have been developed for panning librariesagainst whole cells. These techniques are useful for proteintargets that cannot be easily expressed and purified or pre-sented in an active from. Alternatively, libraries can be direc-ted towards specific cell types to identify particular antigensthat differentiate them. For example, by panning against acancer cell line, a cell-specific marker may be identified. Insuch an instance, the selected target is only relevant in thatit may distinguish a specific cell line from other types of cells.This approach may have particular utility in identifyingtumor-specific antigens. Ligands or antibodies directedtoward these antigens may in turn be armed with drugs forspecific delivery to these cell types.

The ability to pan a particular target expressed on a cellsurface can be significantly more challenging than panningagainst a purified target. Because of the vast heterogeneitypresented on the cell surface and the generally low target con-centration available, elaborate methods are typically requiredto reduce background binding and improve enrichment. Meth-ods to overcome these limitations have relied on over expressionof the receptor as a means of biasing the selection, as well asadditional steps to ensure selectivity. Direct selection againsta specific target on cells has only been successful in a few casesand these have involved libraries biased towards the target

152 Dennis

(14,15). Ligand elution can be used to selectively identify phagethat bind to a specific cell target (16).

Subtractive panning is the most common method forwhich there are a number of variations. Generally, phagelibraries are preincubated with similar cells lacking thereceptor of interest in order to remove nonspecific bindingphage from the library. Rotation between multiple cell linesexpressing the receptor from one round to the next, thusensuring that the background is constantly changing, canfacilitate the selection of specific clones (17,18). For example,De Lorenzo et al. (18) rotated their selection against two dif-ferent target expressing cell lines in the presence of the samecells lacking target to attain clones that bound specifically toErbB2 from an scFv phage library. Phage bound to fluores-cently labeled target bearing cells were separated from themixture by FACS.

In comparison to selecting against specific targetsexpressed on cells, the ability of phage libraries to identifyantigens that distinguish between cell types has been muchmore successful through competitive selections to preabsorbor competitively absorb undesired phage (19–23). Alterna-tively, while selecting against a target receptor presented onimmobilized cells, the same cell type lacking the targetedreceptor can be presented in suspension (Fig. 2). These meth-ods tend to competitively eliminate nonselective phage andallow for the selection of clones that bind to distinguishingregions on the cell surface.

The success of cell selection strategies can depend uponthe method used for separating bound and unbound phage.In addition to washing directly immobilized target cells, cen-trifugation through a density gradient (24) or nonmiscibleorganic phase (22) has been used to separate cells and boundphage from unbound phage (Fig. 3). These methods rapidlyremove nonspecific phage, providing lower backgrounds andhigher cell bound phage recoveries than traditional washingof adherent cells.

Following three rounds of this selective process,Giordano et al. (22) found that 67% of clones from a polyvalentpeptide library selected against vascular endothelial growth


factor-stimulated human endothelial cells displayed enrich-ments of less than 10-fold, yet when compared for bindingto directly immobilized receptor enrichments of up to 1000-fold were observed. This illustrates that while lower back-grounds and higher specific phage recoveries can be obtainedin comparison to traditional washing of adherent cells, theability to identify rare or lower affinity clones may still beexpected to be much more challenging using cells as opposedto standard target immobilization methods outlined above.

To separate phage bound to target cells from competingnontarget cells, a number of methods including FACS andselective capture (biotinylation) have been used. For example,the separation of phage bound to a mixture of particular celltypes, each distinguished by fluorescently labeled antibodiesdirected to cell-specific markers, can be accomplished usingFACS (25). A selectable marker such as random biotinylationof the cell surface can identify cells bearing the target anti-gen, and target cells can be mixed with nontarget cells for

Figure 2 Selective binding of phage to target cells by competitiveselection. The phage library is mixed with nontarget cells in suspen-sion. The mixture is added to target cells immobilized on plastic.The nontarget cells adsorb phage library members that recognizeepitopes present on both cells types. Only unique targets on theimmobilized cells recruit phage binding.

154 Dennis

selection in solution followed by specific capture of targetcells (26).

The pathfinder approach offers a novel and highly selec-tive means of identifying phage that bind to a cell-surfacereceptor. This approach utilizes a ligand or antibody (thepathfinder) to the receptor conjugated to horseradish peroxi-dase (HRP), in order to colocalize HRP in the vicinity of thetarget (Fig. 4A and B) (27). Horseradish peroxidase is thenused to catalyze the conversion of biotin tyramine (BT) to afree radical that will biotinylate the nearest neighboring

Figure 3 Phage separation by selective centrifugation. The phagelibrary is mixed with cells in suspension. Cells, along with boundphage, are separated by centrifugation. Nonprecipitated phagecan be used in subtractive panning. Alternatively, phage associatedwith centrifuged cells can be isolated.


nucleophile. Target bound phage are biotinylated and can bepreferentially recovered from nonspecific, nonbiotinylatedphage using streptavidin-coated magnetic beads (27). Thus,the method offers the ability to select for phage binding to epi-topes neighboring existing ligand-binding sites. Alternatively,ligands identified in this manner can be used as pathfindersthemselves to drive selection of new ligands to the originalpathfinder site (Fig. 4C and D) (28). Frequently, a pathfinderto deliver HRP to the vicinity of the target is unavailable.In such instances, the addition of a flag tag to the target

Figure 4 The pathfinder approach. (A) Phage are bound to thesurface of a cell expressing target antigen (T) in the presence ofan HRP-conjugated antibody or natural ligand directed to the tar-get. (B) In the presence of hydrogen peroxide, biotin tyramine (�)is covalently linked to phage binding in the vicinity of the HRP. Bio-tinylated phage can then be eluted and recovered using streptavi-din-coated beads. These phage likely recognize the target antigenor the ligand–target antigen complex. (C) Additionally, the elutedbiotinylated phage can be added to fresh target presenting cells inthe presence of streptavidin–HRP. (D) A new aliquot of the phagelibrary is added and a second biotinylation reaction is performed.All biotinylated phage are eluted from the cells, recovered usingstreptavidin-coated beads and screened for the ability to bind tothe original pathfinder site.

156 Dennis

could potentially allow generation of an artificial pathfinderdelivery system.

For the targeted delivery of therapeutics to cells, theability of ligands to be internalized into cells is an importantfeature. This method relies on the ability of phage bound tocellular receptors to be internalized and illustrates many ofthe challenges in panning against cell-surface antigens. Bystringently stripping phage from the cell surface causing onlyinternalize phage to be recovered, this method has the addedpotential advantage of lowering nonspecific backgroundphage (29–31). Becerril et al. (29) panned against cells thatoverexpress ErbB2 and found that recovery of internalizedphage could provide an additional (�10-fold) enrichmentwhen compared to recovery of phage from the cell surface.In an another example, Heitner et al. (30) have used this tech-nique to identify internalizing phage sorted against cellsexpressing the epidermal growth factor receptor (EGFR). Inthis example, both adherent (CHO=EGFR) and nonadherent(A431) cell lines expressing target receptor were used in selec-tions incorporating two styles of subtractive panning. For theadherent cell line, untransfected CHO cells were added insolution to selectively remove the non-target-directed phage.For the nonadherent cell line, the phage library was prese-lected against fibroblasts prior to exposure to target bearingcells. In both cases, binding was performed at 4�C, the cellswere washed and internalization was initiated by warmingthe cells to 37�C. Following an extensive wash, the cells werelysed with triethylamine and recovered phage were amplifiedfor the next round of panning. Increased phage titers wereobserved with successive rounds of panning. After threerounds, individual clones were selected and screened for bind-ing to the EGFR. Although target-specific clones were identi-fied, only 10% of the phage recovered from nonadherent and1% of the phage recovered from the adherent cells was targetspecific. The success of this approach requires that phagebinding the receptor of interest are preferentially internalizedover other cell-surface binding phage, and factors such as thetarget receptor expression level and rate of internalizationrelative to other competing receptors are likely important.


Despite this clever combination of subtractive panning andselection, these examples point to the difficulties involved inpanning whole cells.

An interesting enhancement of the selection for interna-lizing phage involves the inclusion of a reporter gene such asGFP under the control of a mammalian promoter on the pha-gemid. Upon internalization, transcription and translation ofthe reporter gene allows selective sorting by FACS and recov-ery of the internalized phage (32).

IV.C. Integral Membrane Targets

The extracellular domains of many cell membrane receptorscannot be expressed and purified to allow selection againsta purified target. For example, this approach is not feasiblefor multitransmembrane spanning receptors such as theG-coupled receptors. These receptors represent a large andimportant family of proteins with a proven history of beingexcellent drug targets. Because of their integral membraneassociation, most of these receptors cannot be easily purifiedand they have not been pursued with phage-displayedlibraries.

One approach for selecting against integral membranetargets is to present them in liposomes. For example, theenvelope glycoproteins from HIV-1 were first solubilizedusing a nonionic detergent and then reconstituted into vesi-cles by the removal of detergent using gel filtration (33).The vesicles were then captured for panning in microtiterplates using a lectin known to bind the envelope glycopro-teins. Potentially, many other detergent solubilized mem-brane proteins could be constituted into liposomes and usedfor phage selections as well.

A method that captures these receptors properly ori-ented in their native lipid environment as liposome-encasedparamagnetic beads has recently been demonstrated (34).Paramagnetic beads coated with streptavidin and a flagtag-directed antibody are used to capture flag-tagged re-ceptor from cell lysates. Detergent-solubilized lipid contain-ing 0.1–1% biotinyl-DOPE is added, and upon dialysis to

158 Dennis

remove the detergent, a lipid bilayer assembles around thebead and provides a native environment for the receptor(Fig. 5). Further studies have found, however, that thebiotin–streptavidin bridge between the bead and the lipid

Figure 5 Formation of paramagnetic CCR5 proteoliposomes. Thesurface of nonporous paramagnetic beads was covalently conjugatedwith streptavidin and an antibody that recognizes the C-terminal C9tag on CCR5. The conjugated beads were used to capture theC9-tagged CCR5 from the cell lysate. After extensive washing, thebeads were mixed with detergent-solubilized lipid containing0.1–1% of biotinyl-DOPE. During the removal of detergent by dialy-sis, the lipid bilayer membrane self-assembles around the beads andCCR5 is returned to its native environment.


bilayer is not essential, thus simplifying generation of theparamagnetic proteoliposome (35). Following exposure ofthese beads to phage-displayed libraries, a magnetic separa-tor is used to capture receptor specific phage. Putative recep-tor-binding phage clones can be analyzed by FACS forbinding to the receptor expressed in cell culture, and thiscan also be used to confirm the integrity of the receptor dis-played on paramagnetic liposomes. Aside from providingpure receptor in a native form, the concentration of receptorin liposomes is considerably increased relative to the levelstypically expressed on cells. Additionally, the flag tag pro-vides selective orientation of the receptor and gives thismethod an advantage over standard liposomes.

A method that combines competitive cell selection andselective sedimentation techniques mentioned above involvedscreening of a scFv phage library against thymic tissue frag-ments in the presence of lymphocytes and spleen cells (23).Tissue fragments were allowed to sediment along with boundphage, while unbound and lymphocyte or spleen cell asso-ciated phage were removed. Isolated phage clones displayedscFvs that selectively bound to the thymic tissue fragmentsand were used to identify specific thymic stromal markers.

IV.D. In Vivo Selections

An ingenious way to identify phage ligands capable of homingto a specific cell type or tissue is to actually inject phagelibraries into live animals (36). Nonspecific phage are distrib-uted and absorbed to surfaces throughout the entire animalwhile harvesting the tissue of interest can preferentiallyretrieve selective targeting ligands. This technology does notrely on binding to a known target, but rather upon the localexpression of unidentified targets to achieve selectivity. Thephage-derived ligands can in turn be used to identify thesemarkers that can provide endothelial targets useful forselective delivery of drugs to specific tissues (37). It is possiblethat these same targets provide the homing signals used bytumor cells and leukocytes to target-specific organsand tissues.Amap formany of these endothelialmarkers has recently been

160 Dennis

described for humans using a phage-displayed peptide libraryinjected into a patient meeting the formal definition of brain-based determination for death (38). Peptide motifs thatprovided specific homing to bone marrow, fatty tissue, skeletalmuscle, prostate, and skin were identified and suggest poten-tial candidate proteins that may act on these tissues. As tumortargeting technology progresses, the relatively short in vivohalf-life of peptides may make tumor selective targetingpeptides useful delivery vehicles for radioimmunotherapy (39).

REFERENCES

1. Scott JK, Smith GP. Searching for peptide ligands with anepitope library. Science 1990; 249(4967):389–390.

2. Bass S, Greene R, Wells JA. Hormone phage: an enrichmentmethod for variant proteins with altered binding properties.Proteins 1990; 8:309–314.

3. Dennis MS, Lazarus RA. Kunitz domain inhibitors of tissuefactor—factor VIIa. II. Potent and specific inhibitors bycompetitive phage selection. J Biol Chem 1994; 269(35):22137–22144.

4. Cunningham BC, et al. Production of an atrial natriureticpeptide variant that is specific for type A receptor. EMBO J1994; 13(11):2508–2515.

5. Li B, et al. Receptor-selective variants of human vascularendothelial growth factor. Generation and characterization.J Biol Chem 2000; 275 (38) :29823–29828.

6. Wrighton NC, et al. Small peptides as potent mimetics of theprotein hormone erythropoietin. Science 1996; 273(Jul 26):458–463.

7. Krebs B, et al. High-throughput generation and engineeringof recombinant human antibodies. J Immunol Methods 2001;254:67–84.

8. Deshayes K, et al. Rapid identification of small binding motifswith high-throughput phage display: discovery of peptidicantagonists of IGF-1 function. Chem Biol 2002; 9:495–505.


9. Kirkham PM, Neri D, Winter G. Towards the design of anantibody that recognises a given protein epitope. J Mol Biol1992; 285(3):909–915.

10. Dennis MS, Lazarus RA. Kunitz domain inhibitors of tissuefactor—factor VIIa. I. Potent inhibitors selected from librariesby phage display. J Biol Chem 1994; 269(35):22129–22136.

11. Devlin JJ, Panganiban LC, Devlin PE. Science 1990; 249:404–406.

12. Lam KS, et al. Nature 1991; 354:82–86.

13. Hawkins RE, Russell SJ, Winter G. Selection of phage anti-bodies by binding affinity mimicking affinity maturation.J Mol Biol 1992; 226:889–896.

14. Portolano S, McLachlan SM, Rapoport B. High affinity, thyr-oid-specific human autoantibodies displayed on the surface offilamentous phage use V genes similar to other autoantibodies.J Immunol 1993; 151:2839–2851.

15. Szardenings M, et al. Phage display selection on whole cellsyields a peptide specific for melanocortin receptor 1. J BiolChem 1997; 272(44):27943–27948.

16. Doorbar J, Winter G. Isolation of a peptide antagonist to thethrombin receptor using phage display. J Mol Biol 1994;244:361–369.

17. Goodson RJ, et al. High-affinity urokinase receptor antago-nists identified with bacteriophage display. Proc Natl AcadSci USA 1994; 91:7129–7133.

18. De Lorenzo C, et al. A new human antitumor immunoreagentspecific for ErbB2. Clin Cancer Res 2002; 8(June):1710–1719.

19. Marks JD, et al. Human antibody fragments specific forhuman blood group antigens from a phage display library.Bio=Technology 1993; 11(October):1145–1149.

20. Noronha EJ, et al. Limited diversity of human scFv fragmentsisolated by panning a synthetic phage-display scFv librarywith cultured human melanoma cells. J Immunol 1998;161:2968–2976.

162 Dennis

21. Ridgway JBB, et al. Identification of a human anti-CD55 sin-gle-chain Fv by subtractive panning of a phage library usingtumor and nontumor cell lines. Caner Res 1999; 59:2718–2723.

22. Giordano RJ, et al. Biopanning and rapid analysis of selectiveinteractive ligands. Nat Med 2001; 7(11):1249–1253.

23. Van Ewijk W, et al. Subtractive isolation of phage-displayedsingle-chain antibodies to thymic stromal cells by using in-tact thymic fragments. Proc Natl Acad Sci USA 1997; 94:3903–3908.

24. Williams BR, Sharon J. Polyclonal anti-colorectal cancer Fabphage display library selected in one round using density gra-dient centrifugation to separate antigen-bound and free phage.Immunol Lett 2002; 81:141–148.

25. de Kruif J, et al. Rapid selection of cell subpopulation-specifichuman monoclonal antibodies from a synthetic phage antibodylibrary. Proc Natl Acad Sci USA 1995; 92:3938–3942.

26. Siegel DL, et al. Isolation of cell surface-specific human mono-clonal antibodies using phage display and magnetically-activated cell sorting: applications in immunohematology.J Immunol Methods 1997; 206:73–85.

27. Osborn JK, et al. Pathfinder selection: in situ isolation of novelantibodies. Immunotechnology 1998; 3:293–302.

28. Osborn JK, et al. Directed selection of MIP-1a neutralizingCCR5 antibodies from a phage display human antibodylibrary. Nat Biotechnol 1998; 16:778–781.

29. Becerril B, Poul M-A, Marks JD. Toward selection of interna-lizing antibodies from phage libraries. Biochem Biophys ResCommun 1999; 255:386–393.

30. Heitner T, et al. Selection of cell binding and internalizing epi-dermal growth factor receptor antibodies from a phage displaylibrary. J Immunol Methods 2001; 248:17–30.

31. Barry MA, Dower WJ, Johnston SA. Toward cell-targetinggene therapy vectors: selection of cell-binding peptides fromrandom peptide-presenting phage libraries. Nat Med 1996;2(3):299–305.


32. Poul M-A, Marks JD. Targeted gene delivery to mammaliancells by filamentous bacteriophage. J Mol Biol 1999; 288:203–211.

33. Labrijin AF, et al. Novel strategy for the selection of humanrecombinant Fab fragments to membrane proteins from aphage-display library. J Immunol Methods 2002; 261:37–48.

34. Mirzabekov T, et al. Paramagnetic proteoliposomes containinga pure, native, and oriented seven-transmembrane segmentprotein, CCR5. Nat Biotechnol 2000; 18(June):649–654.

35. Babcock GJ, et al. Ligand binding characteristics of CXCR4incorporated into paramagnetic proteoliposomes. J Biol Chem2001; 276(42):38433–38440.

36. Pasqualini R, Ruoslahti E. Organ targeting in vivo usingphage display peptide libraries. Nature 1996; 380(Mar 28):364–366.

37. Rajotte D, Ruoslahti E. Membrane dipeptidase is the receptorfor a lung-targeting peptide identified by in vivo phage display.J Biol Chem 1999; 274(17):11593–11598.

38. Arap W, et al. Steps toward mapping the human vasculatureby phage display. Nat Med 2002; 8(2):121–127.

39. Kennel SJ, et al. Labeling and distribution of linear peptidesidentified using in vivo phage display selection for tumors.Nucl Med Biol 2000; 27:815–825.

164 Dennis

5

Phage Libraries for DevelopingAntibody-Targeted Diagnostics and

Vaccines

NIENKE E. VAN HOUTEN and JAMIE K. SCOTT

Department of Molecular Biology andBiochemistry, Simon Fraser University,

Burnaby, British Columbia, Canada

I. INTRODUCTION

Phage library technology is a powerful tool for targeting anti-bodies (Abs) of known or unknown specificity. In this chapter,we examine techniques for applying this technology toAb-dependent applications, specifically, epitope mappingand the development of diagnostics and vaccines. Abs pro-duced during bacterial and viral infections, allergic responsesor chronic illnesses, such as autoimmune disease and cancer,can provide information about the immune response, and

165

sometimes, disease etiology. A proportion of serum Abs maybe biologically active. For example, some may neutralize virusor block bacterial infection, whereas others may trigger celllysis by complement or mediate Ab-dependent cellularcytotoxicity (ADCC). Phage display libraries can be screenedwith Abs to map epitopes on protein antigens (Agns) and tostudy ligand interactions. Moreover, ligands for biologicallyactive Abs can serve as candidate leads for therapeutic andvaccine development.

By screening phage libraries with an Ab or mix of Abs,highly specific ligands can be identified. Different types ofphage display libraries are available for a diversity of applica-tions. Whole-antigen (Agn) libraries, encoded by full-lengthcDNA, can be used to identify proteins that bind to an Ab orserum of interest. Agn-fragment libraries (AFLs), producedfrom fragmented cDNAs, can be used to identify subregionsof a protein that bind Ab. Random peptide libraries (RPLs),made from degenerate oligonucleotides, can identify ligandsfor a wide variety of Abs. In addition to protein-binding Abs,RPLs can be used to identify peptide ligands for Abs that bindnon-protein Agns, such as carbohydrate (CHO) or DNA.Below, we discuss the molecular basis of Ab–ligand interac-tions in the context of phage library screening.

Agn is defined as ‘‘any molecule that can bind specificallyto an Ab’’ (1). Agns include proteins, CHOs, DNA, lipids, and avariety of haptens; the latter are small molecules that, ontheir own, are not immunogenic, but can be made immuno-genic by being chemically coupled to an immunogenic carrierprotein. Organisms and cells can be Agns, including infec-tious organisms, plant and animal components, and evencomponents of oneself (self-Agns). Abs bind specific regionsof an Agn at a site referred to as the antigenic determinantor epitope. Conversely, the Agn-binding site on an Ab isreferred to as the Ab combining site or paratope. Typically,there are many unique epitopes on protein Agns, whereasfewer unique epitopes exist on Agns composed of repeatingunits, such as CHO or DNA. Haptens, being very small andtypically buried within an Ab paratope, comprise a singleepitope. The term Agn is not synonymous with immunogen.

166 van Houten and Scott

An immunogen is an Agn that can elicit specific Ab; however,not all Agns are immunogenic. For example, self-Agns, thatare non-immunogenic in healthy individuals, elicit Ab in someautoimmune diseases and cancers. Anti-self Abs may also beintentionally produced by immunizing with self-Agn mixedwith an adjuvant and=or by immunizing with the self-Agnchemically coupled to an immunogenic carrier protein.

An Ab that is produced by a single B-cell clone and isexpressed from a single set of heavy and light chain genesis said to be a monoclonal antibody (MAb). Each MAb bindsa single epitope on the corresponding immunogen; in addi-tion, some MAbs may have multiple reactivities (i.e., theybind to more than one Agn). Screening libraries with aMAb will result in a restricted set of ligands that bind thewhole paratope or a subsite on it. Thus, ligands that bindan Ab functionally mimic the Ab-binding epitope on the corre-sponding, cognate Agn. In some cases, these ligands can beused as immunogens to elicit the production of specific Abin vivo.

The serum Ab response to immunization comprisespolyclonal antibodies (PCAbs), whose diversity depends uponthe complexity of the immunogen. PCAb responses consist ofmultiple Abs that target a variety of epitopes on the Agn sur-face. Typically, oligoclonal Ab responses are elicited by immu-nization with Agns having a limited number of epitopes(haptens, DNA, lipids, and CHOs); such Ab responses aresometimes genetically restricted to a few VH–VL gene combi-nations. In contrast, whole proteins and organisms (e.g.,viruses and bacteria) elicit highly diverse and complex PCAbresponses. PCAbs typically isolate multiple sets of relatedpeptides or AFs (i.e., peptides or AFs that share sequencemotifs or sequences, respectively) from RPLs and AFLs. Eachset of related clones is specific for a single Ab reactivity, suchthat clones sharing related sequences will compete with eachother in binding to an identical subset of PCAbs, whereasclones bearing unrelated sequences will not. It is importantto note that there is bias in the type of peptide a PCAb willselect from an RPL; peptides that cross-react best withepitopes on the immunogen will emerge as dominant, even

Phage Libraries for Diagnostics and Vaccines 167

though the Abs that select them may not be the most domi-nant in the serum Ab response (2,3). Similarly, the AFsselected from AFLs by PCAbs will be those that best mimicepitopes on the folded protein, and therefore, cross-react withthe PCAb; again, these may not correspond to the most immu-nodominant epitopes.

Protein epitopes bind Ab via specific residues that contri-bute to binding affinity by making low- and high-energy con-tacts with the Ab paratope. Critical-binding residues (CBRs)in a protein epitope make high-energy contacts with the Abparatope. Amino acid replacements can be used to identifyCBRs on a protein, but it must be demonstrated that theydo not affect the global protein fold (4). Protein epitopes canbe described in a number of ways. Conformational epitopesrely on protein folding for binding activity, and Agn denatura-tion destroys Ab binding to these sites. Linear epitopes (alsocalled continuous epitopes) consist of CBRs that are closetogether in a short polypeptide sequence. They are often resis-tant to denaturation, and most of them have been defined bycross-reactivity with peptides (see below). In contrast, discon-tinuous epitopes result from protein folding that bringsdistant CBRs close together (5); these are typically conforma-tional epitopes. Linear epitopes can sometimes have a confor-mational component, with Ab binding being ablated by Agndenaturation, especially reduction of disulfide bridging. Inthe Ab response against folded protein Agn, Abs againstconformational (mainlydiscontinuous) epitopes typically domi-nate, as evidenced by a large decrease in serum Ab bindingto denatured protein Agn.

Peptide mimicry is defined by ligand interactions withthe Ab paratope. Functional mimics of an epitope are cross-reactive, in that they compete with cognate Agn for bindingto Ab. CBRs in peptide mimics are defined by amino acidreplacements that strongly reduce affinity (6,7). PeptideCBRs can mediate binding either through direct, high-energycontacts with the paratope, or by promoting a conformation ofthe epitope that will allow these contacts to be made. Struc-tural mimics both cross-react with Agn and make the samecontacts with the Ab paratope as the cognate epitope; they


may or may not have the same contact residues or structureas the cognate epitope. Structural mimic peptides are moreapt to mimic linear epitopes than discontinuous ones. (Theword ‘‘mimotope’’ was originally coined to refer to structuralpeptide mimics of discontinuous epitopes, but has becomesynonymous with functional peptide mimics.) Immunogenicmimics of an epitope are peptides that, when used as immuno-gens, elicit Abs that bind the cognate epitope. Although a pep-tide probably has to be a functional mimic of its cognateepitope to be an immunogenic mimic, it is not clear that it alsohas to be a structural mimic.

In general, peptides isolated from different types ofphage display libraries bind Abs by different mechanisms.Peptides derived from AFLs with anti-protein Abs bind by amechanism similar or identical to the cognate epitope. Inmany circumstances, these will be linear epitopes or discon-tinuous epitopes within a small domain, as many epitopedeterminants for discontinuous epitopes may be missing froman AFL. Ligands from RPL screening can mimic the functionand=or structure of native protein epitope. Typically, thebinding mechanism for functional mimics of discontinuousepitopes is mixed, with some CBRs making identical contactsto the native epitope, and others promoting binding to the Abparatope by entirely different mechanisms. It has been ourobservation that peptides that mimic discontinuous epitopesalmost always require constraints imposed by disulfide brid-ging. In contrast, linear epitope mimics require structuralconstraints found in the cognate epitope (e.g., beta-turn struc-tures), and may not require constraints by disulfide bridging(see Ref. 2). Thus, the mechanism of protein-epitope mimicryvaries, depending on the type of epitope an Ab binds and thetype of library screened.

Peptides can be functional, and sometimes structural,mimics of CHO and other non-protein Agns. Examples offunctional, but probably not structural, CHO mimicry weredescribed by Harris et al. (8), who screened 11 RPLs with 3very similar MAbs against the cell wall polysaccharide ofgroup A streptococcus. Mapping of the cognate CHO epitopewith a panel of synthetic oligosaccharides showed that the


MAbs recognize the same CHO epitope; moreover, theseMAbs have closely related gene sequences. Despite the simi-larity of the MAbs, the consensus sequences of the peptidesselected by each MAb were distinct, with most peptides beingspecific for only the selecting MAb, and rarely cross-reactivewith the other, closely related MAbs. These results suggestthat the mechanisms by which the MAbs bind to cross-reactive peptides and to CHO differ: peptide binding is pri-marily determined by unique features of each Ab, whereasCHO binding is determined by features that are sharedamong the MAbs. This concept has been further supportedby structural studies by Vyas et al. (9,10).

The fundamental uses of phage display libraries withAbs are to identify the cognate protein Agn for an Ab, tomap an epitope (if the Agn is protein), or to simply identifyligands that cross-react with an Ab. These ligands can be usedas tools for probing an Ab and=or its cognate epitope, asdiagnostics that will recognize the presence of Abs againstthe cognate Agn, and as vaccine leads that will elicit Abs hav-ing the same specificity as the screening Ab. Different types oflibrary are suited for each purpose. Whole-Agn libraries arebest for identifying the cognate protein Agn for an Ab. AFLsare used to map linear epitopes on a protein, or discontinuousepitopes that are located within a small, folded domain. RPLsare useful for identifying ligands that cross-react with Agn;these ligands may or may not be mimics of the native epitope.In the following section, we review in more detail the generaluses and optimal application of each library type, focusing onthe kinds of information one can expect to obtain from eachlibrary type.

II. PHAGE-DISPLAY LIBRARIES AS TOOLS FOREPITOPE DISCOVERY

II.A. Types of Display

The proteins and peptides displayed by filamentous phagelibraries are typically fused to the gene 3 minor coat protein,pIII, or to the gene 8 major coat protein, pVIII. Three types of


display have been described for applications ranging fromwhole protein to peptide display (11). One type of display com-prises the so-called Type 3 and Type 8 vectors, which fuse thelibrary amino acid sequence to the N-terminus of all copies ofpIII or pVIII, respectively. Thus, gene 3 or gene 8 in the phagegenome is modified to encode a library. There is little restric-tion to the variety of proteins and peptides that can be dis-played by Type 3 vectors, probably due to the relatively lowproduction of pIII and low copy number of this molecule onthe virion. In contrast, Type 8 vectors display peptides fusedto each copy of pVIII along the entire length of the phagebody, such that pVIII fusions are expressed at much higherlevels and comprise the entire length of the virion. As a con-sequence, Type 8 libraries are restricted to peptides of fiveresidues or less.

In contrast, the Type 33=88 and Type 3þ3=8þ8 systemsproduce ‘‘hybrid’’ virions that bear two forms of the pIII orpVIII protein: recombinant coat protein fusion and wild-type(WT) coat protein. There are advantages to both pIII- andpVIII-based hybrid systems. Type 3þ3 phagemid systemsare especially useful for cloning more complex proteins (e.g.,for Fab display), whereas Type 8þ8 or 88 systems allowlonger peptides to be displayed. Type 33 and Type 88 vectorsencode both recombinant and WT copies of the coat proteingene on the viral genome, whereas the Type 8þ8 and 3þ3systems use phagemid to express coat protein fusions andhelper phage to express the WT coat protein. For example, aType 88 vector, such as f88–4 (12), is a phage that carrieson its genome a WT gene 8 along with an expression cassettefor recombinant gene 8 ligated to DNA encoding a library.This produces phage with coats that contain both WT pVIIIand recombinant pVIII fusion, and the levels of recombinantpVIII incorporation depend upon the sequence of the displayedpeptide and the phage display system used. For Type 3þ3 and8þ8 systems, phagemid is used, which carries an f1 phage ori-gin of replication and an expression cassette encoding thelibrary protein or peptide fused to coat protein (pIII or pVIII);it also contains a selectable marker such as beta-lactamase.Phagemid DNA is used to transform Escherichia coli, which


are then super-infected with helper phage to provide the otherproteins necessary for phage assembly, including WT pVIII orpIII. Thus phagemid DNA is selectively packaged into virionscomprising library-pVIII fusions expressed by the phagemidand all other proteins provided by helper phage (11). For anextensive review of phage display vectors, library constructionand screening, refer toBarbas et al. (13). The type of display canaffect the screening outcome, as discussed at the end of thissection.

II.B. Whole-Agn Libraries

Phage libraries made from cDNA display the protein reper-toire of an organism or cell line in full-length form. Suchwhole-protein Agns are more likely to be displayed in theircorrectly folded state if they do not require post-translationalmodifications, such as glycosylation; such proteins are typi-cally difficult to express in E. coli. Crameri and Suter (14) firstdescribed the pJuFo vector for a whole-Agn library. ThepJuFo vector system was designed to overcome problemsassociated with the presence of stop codons at the 30 ends ofcDNAs. Since the C-terminus of pIII is believed to be buriedin the viral capsid, displayed proteins must be fused to theN-terminus of pIII for efficient display. Thus, for display onpIII, the 30 end of a cDNA must be ligated to the 50 end ofthe gene 3 coding region. This presents two problems in mak-ing a cDNA display library; the presence of stop codons at the30 ends of cDNAs will prevent protein fusions from beingmade, and it is not possible to predict where open readingframes (ORFs) start for all cDNAs, especially those encodinguncharacterized proteins. Expression of the cDNA as a pro-tein, independent from pIII expression, circumvents thiscomplication. To construct libraries expressed by the pJuFophagemid vector, whole proteins encoded by cDNA fragmentsare fused to the C-terminus of the leucine-zipper region of theFos transcription factor. The same vector also encodes theleucine zipper region of Jun fused to the N-terminus of pIII.Phage assembly results in heterodimerization between theFos and Jun leucine zippers, and cysteines that have been


added to the N- and C-termini of the Fos and Jun leucinezippers form a covalent bond that allows display of the cDNAgene product. Helper phages are required for phage packa-ging and virion assembly. Alternative systems for cDNAdisplay have been described; for example, fusion to theC-terminus of the pVI minor protein overcome some of the dif-ficulties described above (15,16). Others have created systemsfor fusing cDNA to both pVIII and pIII (17).

Whole-Agn libraries have multiple applications, with amajor one being in allergy research. Immunoglobulin E(IgE) from allergic patients is used to identify protein aller-gens from cDNA libraries made from the mRNA of wholeorganisms. Once identified, these proteins can then be usedin clinical diagnostic allergy tests. Whole-Agn libraries aregenerally useful for the identification of unknown Agn. Thistype of library has also been used to identify tumor-specificAgns from cell lines (see below) (18,19). Since this type of dis-play uses a bacterial expression system, problems with properglycosylation and protein folding can arise. To circumventthese problems, whole-Agn display on yeast or mammaliancells could be considered.

II.C. Agn Fragment Libraries

Also referred to as gene fragment libraries or natural peptidelibraries, AFLs are encoded by randomly sheared fragmentsof cDNA rather than complete cDNAs. AFLs can be madefrom the whole genomes of lower organisms (e.g., a viralgenome), the cDNA repertoire of a cell line, or the cDNA orgene of a single protein. AFLs are constructed by digestinga genome (from a lower organism that lacks introns), cDNAfrom total RNA of a cell or organism, or cDNA from a geneof interest, with a non-specific endonuclease such as DNaseI (20). After addition of linker DNA to fragment ends, thefragments are ligated into an appropriate phage vector, andproteins related to the gene fragments are displayed on thephage surface. Digestion can control the size of the fragments,and hence, the length of expressed AFs. However, most of theclones in a library do not encode the parental protein


sequence. Since each fragment can be ligated in the wrongreading frame at either end or it may be inverted, only 1out of 18 possible insertions will encode the parent proteinsequence. Zacchi et al. (21) addressed this problem by devis-ing a system to preselect library inserts for ORFs, before clon-ing them into a phage display vector. The pPAO2 vector wasdesigned to ligate gene fragments upstream from a beta-lactamase gene flanked by lox-recombination sites, which isupstream of the phage gene 3. Gene fragments are clonedas fusions to the beta-lactamase gene, and functional fusionsare selected with ampicillin. Phagemid DNA from ampicillinresistant clones is then used to transform cells expressingthe Cre recombinase; this removes the sequences betweenthe gene fragment and the gene 3, causing ORFs to be joinedin frame to the 50 end of the gene 3 coding region. Superinfec-tion of these cells with helper phage allows production of thelibrary with clones displaying expressed ORF sequences. Theauthors used this system to make a library and reported thatall the clones they analyzed contained ORFs, of which 83%were localized to known genes.

There are multiple applications for AFLs. They are wellsuited to mapping epitopes for anti-protein MAbs or PCAbs.The amino acid sequences of clones isolated from an AFL bya MAb should align to identify the Mab-binding epitope.PCAbs against a protein or set of proteins may isolate setsof aligning peptides corresponding to multiple epitopes. Thismethod is most suitable for the isolation of linear epitopes;however, discontinuous epitopes may be isolated if the librarycontains a folded domain bearing CBRs required for MAbbinding. Since the origin of the epitope is the native protein,optimization by building sub-libraries is not necessary; how-ever, doped libraries may be used to optimize binding, andto characterize the epitope in more detail (see Sec. II.F).

II.D. Random Peptide Libraries

The idea of RPLs displayed by filamentous phage was firstproposed by Parmley and Smith (22), and then reported in1990 (23–25). RPLs consist of ‘‘randomized’’ amino acids fused


to pIII or pVIII, and are usually expressed by Type 3, or Type88=8þ8 vectors, respectively. They range in size from 108 to1011 clones, and cover a range of peptide lengths; the peptidescan be ‘‘linear’’ or ‘‘constrained’’ by the introduction of fixeddisulfide bridges. RPLs are encoded by degenerate oligonu-cleotides that are synthesized using a degenerate codon strat-egy (NNK or NNS, in which N represents an equimolarmixture of all four nucleotides, K is an equimolar mixture ofG and T, and S is an equimolar mixture of G and C). NNKand NNS codons each comprise 32 codons that encode all 20amino acids and one amber stop codon. Since there are 32 pos-sible codons, not all of the amino acids are equally repre-sented in the mixture, producing a bias towards certainresidues. There are three codons for Arg, Leu, and Ser, andtwo codons for Val, Pro, Thr, Ala, and Gly; the remainingamino acids are each encoded by a single codon. This biascan be overcome by codon-based oligonucleotide synthesisstrategies (Glen Research, Sterling, VA). This approach usesmixed trinucleotide phosphoramidites, instead of singlenucleotide phosphoramidites, to synthesize degenerate oligo-nucleotides that will encode a library (26). There are severaladvantages to this. First, stop codons are omitted, allowingmore clones in a library to express and display long aminoacid sequences. Second, since single codons can be mixedtogether in any proportion, codon bias can be completely con-trolled. Last, codons that are optimized for expression canpotentially be used for library construction. The methodsdescribed here for producing degenerate oligonucleotides forRPLs are also useful for peptide optimization, as describedin Sec. II.F.

RPLs can be used to identify peptide ligands for MAbsand PCAbs, regardless of the type of Agn they recognize(i.e., CHO, DNA, or protein). RPLs have an advantage overAFLs in that the same RPL can be used for different applica-tions, whereas an AFL is only specific to one protein or organ-ism. As with AFLs, RPLs can be used to map epitopes onprotein Agns. An RPL would be screened with an anti-proteinMAb, and the sequences of peptides from MAb-selected clonesaligned. Peptide sequences selected with a MAb against a


linear epitope often can be aligned into consensus sequences.The consensus residues, having been selected by the MAb, arepotential CBRs, and they can be confirmed as CBRs by aminoacid replacement studies. If the protein is known, its gene canbe searched for matches to the consensus sequence to identifythe epitope. If the gene for the protein is unknown, consensussequences can be used to perform BLAST searches, whosepurpose is to identify potential Agn targets for the Ab. Thislatter application has been useful, particularly in identifyingAgns and the organisms expressing them that may beinvolved in immune responses. However, this method is unre-liable. It can identify leads; but they must be tested furtherfor MAb binding and other characteristics.

II.E. Library Screening to Isolate High-AffinityLigands or a Broad Range of Ligands

Screening conditions affect the type of ligands isolated froma library. These include the type of display used (pVIII vs.pIII), the panning strategy (in-solution vs. solid-phase), theconcentration of the target molecule and preadsorption toremove phage that bind to plastic or reagents used in screen-ing rather than the desired target.

Solid-phase panning uses immobilized Ab to capturephage. The Ab can be immobilized by adsorption to a plateor to beads, or Ab can be captured by immobilized proteinA; Ab can also be biotinylated and then captured on immobi-lized streptavidin. A phage library is allowed to bind to theplate and non-binding phage are removed by washing. Boundphage can be eluted by Ab denaturation in acid, then theeluates can be neutralized and used to infect E. coli cells thatwill amplify the eluted phage. Alternatively, cells can beadded to the plate-bound phage and infected, particularly ifthe phage vector is a Type 8þ8 or Type 88. In solid-phase pan-ning, phage are captured by multiple Abs in the well of amicrotiter plate or on beads; multivalent binding producesan avidity effect that can effectively capture phage bearingeven relatively weak-binding peptides. Thus, phage bearingpeptides covering a whole range of affinities are captured by


solidphase panning. Importantly, this is the most efficientmethod of capturing phage and is generally used in the firstround of panning, when each clone in the library is availableto be captured by Ab.

In contrast, in-solution panning allows more stringentselection, because binding of each Ab (especially if it is inFab form) to a phage-borne peptide is an independent event.This approach is especially useful in screening sub-librariesdescribed above, as a means of optimizing or further charac-terizing a peptide ligand. In-solution panning begins bymixing Ab and phage, and allowing the reactions to reachequilibrium. Subsequently, Ab-phage complexes are capturedby brief exposure to immobilized streptavidin or protein A. AsAb concentration drives the binding of Ab to peptide, thenumber of Abs bound to a given phage will depend upon theKd between the displayed peptide and the Ab. Thus, at lowAb concentrations (near or below the Kd) phage bearingtighter-binding peptides will bind proportionally more Abmolecules than phage bearing weaker-binding peptides; theformer, having more bound Ab per phage, will be moreefficiently captured by immobilized protein A or streptavidin.Thus, clones having tighter-binding peptides will be favoredin selections using low Ab concentrations. The disadvantageof this strategy is that capture of phage out of solution is inef-ficient, and results in lower yields (27). Thus, it is best used onphage that have been enriched by an initial round of solid-phase panning. A range of Ab concentrations should be usedfor in-solution pannings as it is difficult to estimate the bestconcentration of Ab to use beforehand (27). Since librariestypically consist of millions to billions of phage clones, oneshould also consider practices that will reduce phage bindingto all facets of the screening system, including reagents(streptavidin, protein A), beads, tubes, or plates in whichthe pannings are carried out, and parts of the Ab that arenot involved in Agn binding. To decrease background causedby such phage, preadsorption of the library on these reagentsis also recommended.

Before screening, it should be decided if the goal of theexperiment is to obtain high-affinity clones or clones covering


a broad range of affinities and sequences. In-solution screen-ing with MAb or Fab is more likely to isolate clones bearinghigh-affinity peptides, whereas solid-phase screening PCAbsis best used to map multiple epitopes on a protein. Peptidesdisplayed by Type 3 vectors, and probably by Type 8þ8=88ones, are aligned closely enough to allow bivalent binding toa single IgG (28). Multivalent binding to phage can produceavidity effects that mask intrinsic affinity differences betweenpeptides, and thus decrease affinity discrimination; this effectcan be overcome by using Fab for screening. If PCAbs arebeing used to identify multiple peptide ligands, solid-phasescreenings (or in-solution screenings using relatively highAb concentrations) should be performed to isolate a broadrange of phage encompassing both tight and weak binders.Moreover, PCAbs, being derived from serum, can includeAbs with specificities for Agns other than the target antigen.To avoid Abs that are not specific for Agn, it is prudent toscreen libraries with PCAbs that have been affinity purifiedon Agn. This will ensure that the ligands selected from alibrary will very likely cross-react with Agn.

II.F. Sub-Libraries for Optimization of Peptidesand Agn Fragments

Phage display libraries can be used to optimize ligands for Abbinding by two general approaches. The first is most applic-able to ligands isolated from RPLs whose sequences align toproduce a consensus. Sub-libraries can then be designed inwhich the consensus residues are fixed and all other resi-dues surrounding them randomized (29). Alternatively, sub-libraries can be produced that extend a consensus sequenceby placing additional randomized residues on either side ofit. This creates longer peptides, and thus potentially increasesbinding affinity by increasing binding contacts with the Ab(30–32).

If a single ligand is isolated from an RPL, the CBRs willlikely be unknown. A doped library can be made to identifyCBRs and optimize binding. This approach can also beused for peptides derived from AFLs to define CBRs, confirm


epitope length, and optimize binding. Doped libraries of threetypes can be constructed. In a nucleotide-doped library, eachnucleotide used to encode the peptide sequence is mixed withan N nucleotide mixture; this results in a degenerate oligonu-cleotide that is biased toward the sequence encoding thepeptide ligand, but depending on the level of N-nucleotidedoping, each nucleotide in the sequence will have a chanceof being replaced by a different nucleotide. The drawback ofthese doped libraries is that each N-nucleotide-doped codonis biased to express a different subset of amino acids. Thismakes it difficult to discriminate between amino acidsselected as a result of improved affinity from those that areover-represented in the library. An NNK-codon-doped librarycircumvents this problem, but is more difficult to prepare, asit requires oligonucleotide synthesis on parallel columns (33).During synthesis of each codon that is to be doped, the resinon which the oligonucleotide is being synthesized is splitbetween two columns. On one column, the codon from the par-ental peptide sequence is synthesized; on the other, an NNKcodon is synthesized, and after synthesis the resin from thetwo columns is combined. This procedure is repeated for allresidues that are to be doped. More recently, with the com-mercial availability of trinucleotide phosphoramidites (GlenResearch, Sterling, VA), codon-doped libraries of any typecan be made, and probably more easily than multiple-columnsynthesis. Specific trinucleotide phosphoramidites can bemixed with NNK ones or a limited mixture of trinucleotidephosphoramidites to produce degenerate oligonucleotides(26). A doped library is screened with the initial MAb underconditions that will stringently select relatively tight binders,yielding more in-depth binding and sequence information onthe consensus sequence.

III. DIAGNOSTICS

The sections above reviewed Ab–Agn interactions anddiscussed several types of phage libraries as a source ofAb-binding ligands. In this section, we review the application


of phage libraries to diagnostics, focusing on methods thatdetect the presence of specific Abs in the blood. Many diagnos-tic assays use Agn or epitopes to detect the presence of specificAbs in serum, indicating the presence of a pathogen and=orthe existence of a disease state. For example, a serum sampleassayed for the presence of specific IgE against a particularallergen indicates whether an individual has an allergic sen-sitivity. Such assays can also be used to identify Abs that bindviral or bacterial proteins, indicating infection. Theoretically,even non-infectious states may be assessed with assays devel-oped to specifically detect Abs associated with diseases suchas autoimmune disorders and cancer.

Phage display libraries are useful as a source of Agn fordiagnostic assays. A primary use of phage display libraries indiagnostics is to identify protein Agns that react with Absagainst a known organism (with the exception of membranebound proteins, since they are insoluble). For example, awhole-Agn library produced from genomic or cDNA from anallergenic organism can be screened with reactive sera toidentify specific allergen proteins. These allergens can thenbe produced in their recombinant form, and used in clinicaldiagnostics to test other individuals for reactivity to the sameallergen. Similarly, whole-Agn libraries can be screened toidentify protein ligands for infections caused by a knownpathogen. For example, human immunodeficiency virustype-1 (HIV-1) is diagnosed by the presence of serum Absagainst the viral protein p24. In circumstances in which thediagnostic agent is unstable, rare, expensive or difficult topurify, an Agn fragment or peptide may be a viable alterna-tive. RPLs can also be applied to situations in which onewants to distinguish between two closely related species.For example, sera from individuals infected with Chlamydo-phila pneumoniae can cross-react with C. trachomatis andC. psittaci (34). Identification of peptides that differentiatebetween these species can improve diagnosis of infectiousdisease. Specific Abs are often produced in autoimmune disor-ders and cancer, and perhaps in some idiopathic chronic dis-orders (i.e., chronic fatigue syndrome). Peptide libraries canbe screened with sera from affected individuals to identify


ligands associated with disease-related Abs; however, it isdifficult to determine with certainty whether particularpeptides are associated with a specific disease state. The fol-lowing section reviews the applications of phage libraries tothese problems.

III.A. Whole-Agn Libraries for Agn Identification

Soon after describing the pJuFo vector (see above), Crameri etal. (35) applied it to construct a whole-Agn library from thecDNA of the fungus Aspergillus fumigatus. They isolated sin-gle proteins from the library using IgE from individuals withA. fumigatus allergies. Later, they identified one of these pro-teins as a manganese superoxide dismutase that reacted inskin tests in allergic individuals (36). More recently, thisgroup used high-throughput methods to screen a cDNAlibrary from the allergenic fungus Cladosporium herbarumwith IgE from 13 sensitized individuals. They identified theallergen as a hydrophobic component of the cell wall; this isthe first report of a fungal cell wall protein as being a clini-cally relevant allergen (37). Allergenic proteins have beenidentified by screening whole-Agn libraries produced fromseveral organisms including A. alternata, Candida albicans,wheat germ, peanut, and E. coli (38). Significantly, thisapproach identifies allergens that can be expressed recombi-nantly; other allergen preparations may be comparativelydifficult to produce.

Whole-Agn libraries have been used to identify tumor-specific Agns. The cDNA for such libraries has been producedfrom a cancer cell line, such as breast cancer lines T47D andMCF-7 (19) or the colorectal cancer cell line HT-29 (18).Library screening with sera from cancer patients, includingthose who had been immunized with autologous tumor cells,isolated specific clones, whose sequences identified noveltumor-associated Agns. These examples used pVI displaydescribed by Jespers et al. (15) and, more recently, Huftonet al. (16). The tumor Agns identified may serve as candidatesfor tumor vaccination, sero-diagnosis of cancer, prognostic mar-kers, or as probes for monitoring tumor cell-based vaccination


trials. Other groups have used whole-Agn libraries to identifyauto-Agns (39).

III.B. RPLs for Identifying Diagnostic Peptidesfor Known Pathogens

Folgori et al. (40) first described the screening of RPLs withserum PCAbs from pathogen-infected individuals to identifypeptides for use in disease diagnosis. Success is dependenton identifying peptides that bind all or most sera frominfected individuals, while being unreactive with sera fromuninfected individuals. Kouzmitcheva et al. (41) used thistechnique to identify peptides that indicate Lyme disease.They used affinity-purified PCAbs from eight individualsinfected with Borrelia burgdorferi in a screening strategy thatselected each clone with IgG from at least two individuals.They screened a panel of 12 RPLs with various constraints,and isolated 17 peptides that were recognized by all 10 serafrom infected individuals whose sera had not been used inscreening. None of the consensus sequences isolated matchedB. burgdorferi proteins. A similar technique was used byBirch-Machin et al. (42) to identify peptides that bind toequine herpes virus (EHV) anti-sera; however, they reportedhomology of consensus sequences to several EHV proteins.For these peptides to be useful for diagnostics, they must befurther studied and developed. Ultimately, they must reactwith at least 75% of the infected population to be consideredfor commercial assays (41).

Studies identifying diagnostic leads are best performedwith PCAbs rather than MAbs, and preferably with PCAbsfrom more than one individual. There are several reasonsfor this. First, different individuals or species responding tothe same Agn may not make the same Abs, or may not makeAbs against the same epitope. Second, if a MAb is madeagainst a non-immunodominant epitope, then individualsmay have only a small concentration of that MAb in theirsera, which may be difficult to detect. The work of Benguricet al. (43) illustrates this point. Their goal was to identify pep-tides and Abs that would be specific for a MAb against


Mycoplasma capricolum subsp. capripneumoniae (MCCP).They used murine MAb 4.52 to isolate two peptides from a17-mer RPL. Mab 4.52 was also used to immunize chickens;the resulting chicken PCAbs were adsorbed on mouse IgGto remove anti-isotypic and anti-allotypic Abs, and thenaffinity purified on MAb 4.52 to isolate specific Abs againstthe Ab combining site. (These are called anti-idiotype (Id)Abs, anti-Id Abs, see Sec. V.B for more details about Ids). Bothpeptides and the purified anti-Id Ab blocked MAb 4.52binding to MCCP cell lysate in a competition ELISA. How-ever, when goat anti-sera against inactivated MCCP wastested for binding to both the peptides and anti-Id Abs, nobinding was detected. Thus the peptides and anti-Id Ab werespecific for the murine MAb, but not for Abs from a differentspecies that were presumably against the same Agn (thoughthis was not shown). The authors might have had better luckif they had used a PCAb rather than a MAb for library screen-ing, and perhaps an AFL rather than an RPL as a source ofpeptides.

III.C. AFLs and RPLs for Auto-Agn Identificationand Idiopathic Disease Diagnosis

AFLs and RPLs can be applied to the identification of auto-Agns for autoimmune disorders and may eventually lead todiagnostic tools for specific autoimmune disorders. As anexample of this, Fierabracci et al. (44) used RPLs to identifya new auto-Agn that may be associated with the autoimmunedisease, insulin-dependent diabetes mellitus (IDDM, alsoknown as type II or juvenile-onset diabetes). They screenedtwo 9-mer peptide libraries with a single serum sample froma patient with IDDM, and then screened the enriched phagepools with two additional IDDMþ serum samples, isolatingpeptides that bound to Abs from all three serum samples.To ensure that the peptides identified during the screeningwere IDDM specific, the enriched phage pools were furtherscreened with three additional IDDMþ sera and counter-screened with sera from eight healthy people. Five uniqueIDDM-related peptides were identified by testing 70 phage


clones for binding to serum samples from newly diagnosedand long-term IDDM patients, and for lack of binding to serafrom healthy individuals and patients with other autoim-mune disorders. One peptide (CH1p) that showed no homol-ogy to any human protein was detected by 70% of newlydiagnosed IDDM patient sera compared with 10% of healthycontrol sera.

The authors used the CH1p peptide to identify a puta-tive, IDDM-associated auto-Agn. To further test the abilityof the CH1p peptide to behave as an immunogenic mimic ofan epitope on the auto-Agn, the CH1p peptide was used toimmunize rabbits. The resulting anti-CH1p serum (R-anti-CH1p) was used to screen a cDNA expression library fromhuman pancreatic islet cells. Three clones were identifiedand all shared 99% homology with human osteopontin, a mar-ker for late osteoblast differentiation (45). A number of experi-ments were then performed to assess whether the CH1ppeptide is an immunogenic mimic of osteopontin. SDS-PAGEof human islet cell preparations followed by western blottingwith R-anti-CH1p and anti-osteopontin PCAbs indicated thatboth Ab preparations bound to a protein with a molecularweight corresponding to that of osteopontin. R-anti-CH1p alsobound to purified osteopontin in a dot blot, and binding couldbe blocked by preincubation with CH1p peptide. Immunohis-tochemical staining of paraffin-embedded pancreatic sectionswith both R-anti-CH1p and IDDMþ=CH1pþ sera producedsimilar binding patterns to somatostatin-producing alphacells. However, the R-anti-CH1p, but not the IDDMþ=CH1pþ

sera, was blocked from binding to these cells when preincu-bated with CH1p peptide. It is not clear whether theIDDMþ=CH1pþ sera that were tested also had osteopontin-binding activity (if they did not, then it is perhaps under-standable why the peptide did not block their binding to thecells), nor whether binding of any of these Abs would beblocked by osteopontin. One can conclude that that theCH1p peptide can elicit osteopontin-binding Abs in rabbits,and that such reactivities occur in CH1pþ sera from IDDMpatients. However, the relationship between CH1p-bindingand osteopontin-binding reactivities in IDDMþ sera is not clear.


The frequency of osteopontin reactivity in IDDMþ

patient sera was further analyzed by radioimmunoassay(RIA), using recombinant human osteopontin. Serum samplesfrom 146 people were tested: 64 were from newly diagnosedIDDM patients, four were from patients with long-termIDDM, and the remaining serum samples were controls fromhealthy individuals and patients with autoimmune disorders.As mentioned above, 70% of newly diagnosed IDDM serareacted with the CH1p peptide in ELISA; however, only 8%of the tested sera bound to osteopontin by RIA. If the CH1ppeptide is supposed to be a marker for IDDM-relatedAbs against the putative auto-Agn, osteopontin, it is not clearwhy so many IDDM patients with CH1pþ sera areosteopontin�. A central question from these studies thatremains unanswered is whether Abs from different patientsthat bind to CH1p also bind specifically to osteopontin. Thus,these intriguing findings do not clearly show that osteopontinis an auto-Agn associated with IDDM. Moreover, the fre-quency of osteopontin reactive Abs in patient sera indicatesthat, even if osteopontin is an auto-Agn, it may not be a verycommon one. Interestingly, independent studies from thiswork have shown that osteopontin is involved in the vascularcalcification that accompanies diabetes of all types (reviewedin Ref. 46).

Fierabracci et al. (44) developed a novel approach to thediscovery of putative auto-Agns, simply by finding a peptidethat commonly reacted with serum Abs from diseasedpatients. Under the assumption that this peptide couldbehave as an immunogenic mimic of a corresponding auto-Agn, they used the peptide to obtain anti-peptide sera, andin turn, used this anti-serum to identify a putative cognateauto-Agn. However, to prove the connection between a dis-ease state, an Ab-binding peptide and its putative cognateAgn, several criteria should be met. First, serum Abs fromdiseased patients should be affinity purified on the peptideand then tested for binding to the cognate Agn, to show thatdisease-specific Abs cross-react with both Agns. Second, serathat react with a disease-specific peptide should also reactwith the cognate Agn. If this is not the case (i.e., if more sera


bind peptide than cognate Agn), then there may be confound-ing reactivities in the patient sera. For example, the peptidereactivity may initially arise via one Agn but then ‘‘spread’’to the putative cognate Agn; this could account for sera thatreact with peptide but not the putative cognate Agn. Third,if a putative cognate Agn is truly an auto-Agn, one wouldexpect to find signs of ‘‘epitope spreading’’ on the auto-Agn;more disease-specific sera should react with the cognateAgn than with its corresponding disease-specific peptide.Last, as this approach depends on serum Ab responses, theyshould be tested at every step for non-specific reactivity,which can confound experimental results. One way to removesome, but not all, non-specific reactivities in a serum is toincubate it with a complex antigenic mixture, like a bacteriallysate, then to test treated and untreated sera for binding tothe Agns of interest (disease-specific peptide, putative cognateAgn, and unrelated protein controls). One should see a drop inbinding to the unrelated controls, but not in binding to thetargeted peptides and cognate Agns. Given these caveats, thisapproach, using RPLs to discover disease-specific Ab reactiv-ities, may be useful in identifying new Agns involved in avariety of disease states, including autoimmune diseasesand cancer. The hope is that the pathology of some idiopathicdiseases may be clarified through the identification of disease-associated Agns.

RPLs have also been used to identify ligands for the diag-nosis of autoimmune disorders such as Crohn’s disease (CD).By screening an RPL displaying nonamer peptides with serafrom CD patients, Saito et al. (47) identified five CD-relatedpeptides; only two of them shared similar sequences and bind-ing characteristics. Multiple antigenic peptides (MAPs) madefrom four of the peptide sequences were recognized by 52 of 92(56.4%) patient sera tested, but reacted with only 6.2% ofnegative controls. Others have used a similar approach toidentify peptides that specifically bind an anti-cardiolipin MAbin studying homologous, disease-associated anti-cardiolipinAbs in patients with anti-phospholipid syndrome (48).

There are many diseases for which there is no knownassociated pathogen. While the examples above describe


discovery of disease-related ligands, this approach is notalways successful. Our lab has screened RPLs with PCAbsfrom patients diagnosed with Kawasaki disease, a disorderthat occurs in young children and is characterized by suddenonset of a high fever followed by severe vasculitis and cardiacaneurysm. We screened a panel of RPLs with sera from chil-dren with Kawasaki’s disease who had not received IVIGtreatment; but found no peptides that bound patient serabut not that of controls. Another group screened a panel ofRPLs with purified IgG from 23 patients with chronic fatiguesyndrome and identified specific peptides associated with thisillness; however, no difference in phage clones was seen bet-ween patient and control sera (Smith, personal communication,2003).

Diagnostic applications of phage libraries include theidentification of whole Agn, the identification of AFs when adisease-causing organism is known, or the identification ofpeptides associated with autoimmune or chronic disordersand other idiopathic diseases. The approach of screeningwhole-Agn libraries to identify specific allergens has yieldedmuch success. Other applications have met with limited suc-cess, such as the identification of peptide ligands specific forAbs associated with idiopathic diseases. Still, there is the pro-mise that the diagnosis of, or clues to the pathogenesis of,these latter diseases may be revealed by novel approachesusing RPLs. In the following section, we discuss epitope map-ping, one of the more common uses for phage display libraries.

IV. PHAGE LIBRARIES FOR EPITOPE MAPPING

It is useful to describe traditional epitope mapping to beginwith, and the work of Jin et al. (49) is an elegant exampleof this. They used 21 MAbs to fine-map 19 discontinuousand 2 linear epitopes on human growth hormone (huGH).First, gross epitope regions were mapped using 12 mutantsof huGH that each had a different region replaced with thecorresponding region from a homologous protein. Binding by19 of the MAbs was decreased by more than one mutant,


indicating that it involved more than one region on huGH.Moreover, for each of the 19 MAbs, these regions mapped tolocalized patches on the surface of huGH, further supportingthe idea that they form a discontinuous epitope. Only twolinear epitopes were identified (i.e., MAbs whose bindingwas affected by replacement of only one region), reflectingthe dominance of discontinuous epitopes in the murine Abresponse.

Second, fine epitope mapping was completed using Alareplacement of individual residues within localized regionsidentified in the homolog mapping studies. As expected,the CBRs of epitopes were spatially clustered and mostlysurface-exposed residues; but surprisingly, they wererestricted to only a few types of amino acid: Arg, Pro, Glu,Asp, Phe, and Ile. This elegant example of a detailed epitopemapping study serves as a starting point from which to ana-lyze epitope mapping work performed with phage libraries.

AFLs and RPLs have been used to identify epitopes on aprotein Agn for MAbs and PCAbs. Linear epitopes are rela-tively easy to identify for two reasons. First, they are typicallyrestricted to a short sequence, making it likely that peptidescontaining fragments of the appropriate length, or longer, willbe present in an AFL or RPL. Second, linear epitopes usuallycomprise three to five CBRs, most of which are residues thatare shared with the cognate protein Agn. Thus, a consensussequence has a good chance of aligning with the cognate pro-tein’s sequence, even if it originates from an RPL. The advan-tage of an RPL is that CBRs in an epitope are more likely tobe identified from the sequences of Ab-binding clones(through the identification of consensus residues), whereasthe advantage of an AFL is that Ab-binding clones containmost or all of the cognate epitope, whose boundaries aredefined by alignment of sequences from multiple clones.

Discontinuous epitopes are more difficult to identifyusing either type of library. Compared to linear epitopes, dis-continuous epitopes are less likely to be present in an AFL,because of their distantly spaced CBRs and=or their depen-dence on the folding of a larger protein domain, Hence, theytypically require a longer protein sequence. Nevertheless


AFLs are best for identifying a discontinous epitope, given therestrictions that: (i) the targeted epitope must be reproducedby a relatively short region or small domain from the cognateAgn, (ii) an AF containing the discontinuous epitope will mostlikely be considerably larger than the epitope itself, and (iii)CBRs will not be identified by this approach. The smallestAb-binding AF could be used to produce a doped library,and Ab-binding clones coming from this sub-library mayreveal CBRs.

Alternatively, ligands for Abs against discontinuousepitopes can be identified from RPLs. If CBRs are identifiedfrom clones isolated from an RPL, they can be used to mapa discontinuous epitope. However, many of the CBRs mayserve to promote the unique structure of the peptide, withoutcontacting the Ab directly. In this instance, only a few CBRsmay match between a peptide ligand and its correspondingdiscontinuous epitope, making it difficult to identify theepitope by sequence alignment with the cognate Agn. Pep-tides that mimic discontinuous epitopes typically have moreCBRs than peptides that mimic linear continuous epitopes,and they are typically less abundant in a library. Thus, itcan be difficult to isolate multiple, independent clones sharinga consensus sequence by a typical library panning procedure.A rare, discontinuous-epitope mimic derived from an RPLmay require, in addition to the initial RPL screening, theeffort of producing and screening sub-libraries to reveal itsCBRs and to produce acceptable affinity. For these reasons,discontinuous epitopes can be difficult to map regardless ofthe type of library used for Ab screening.

PCAbs have been used to map multiple epitopes on aprotein or organism (2,3,42,50). It is not possible to predictthe Abs in a PCAb mixture that will cross-react with peptidesfrom either an AFL or RPL, and thus a cross-reactive peptidemay not be selected for a single epitope of interest that isbound by a PCAb. PCAbs usually isolate a number of peptidescorresponding to different Ab subspecificities; thus, they canprovide a peptide ‘‘signature’’ that reflects, in part, thebreadth of the Ab response. Moreover, such peptides can beused to compare Ab responses from different individuals


and from such comparisons, to identify immunodominantepitopes that are common to multiple sera (51,52). The useof multiple libraries in PCAb screenings (3,41,53) can providea wider range of selected clones.

Epitope mapping can be used to define epitopes for MAbswith known biological function. By using such MAbs forscreening, AFLs and RPLs have been used to identify bindingsites for neutralizing Abs (50,54,55) or immunodominantregions on Agn (50,56–58). Moreover, peptides have potentialuse in the design of vaccines that specifically target aresponse against a cognate epitope on an antigen, includingpathogens and toxins (see below). Thus, RPLs and AFLs havethe potential for producing vaccines that target particularAbs, and thus, specific epitopes.

IV.A. AFLs for Epitope Mapping

AFLs are used to identify regions of a protein that bear acognate epitope. Very likely, the whole epitope must be pre-sent on a fragment for binding to occur. Once an AFL isscreened, sequences from binding clones can be aligned toreveal the epitope, deduced as the shortest region shared byall binding clones. As the screening of an AFL may not alwaysyield ligands (especially if a MAb binds a discontinuous epi-tope), RPLs may be screened along with the AFL to furtherensure that clones bearing ligand peptides will be isolated.Fack et al. (59) used four MAbs (each against a differentprotein) to screen two RPLs and four AFLs (one AFL for eachprotein). Two of the MAbs (215 and Bp53–11) were known tobind linear epitopes, and the other two MAbs (GDO5 andL13F3) were also thought to bind linear sequences, but theyhad not been previously mapped. The RPLs yielded consensussequences for only two MAbs (Bp53–11 and GDO5). In con-trast, all four MAbs isolated ligands from their respectiveAFLs. Alignment of the sequences isolated by each MAbdetermined the region that was shared between clones, andthis shared sequence was used to define the cognate epitopeon the protein. The two MAbs with known epitopes (215and Bp53–11) selected several short fragments from their


AFLs, confirming their previously identified linear epitopes.One of the MAb with an unknown epitope (GDO5) isolatedthree overlapping AFs, which identified a 10-residue regioncontaining a putative linear epitope. All putative epitopeswere confirmed by testing Ab-binding to overlapping peptidesthat spanned the region identified by the AFs or RPLs. In con-trast, it was concluded that the fourth MAb (L13F3) binds adiscontinuous epitope, since it isolated two partially overlap-ping AFs that spanned 50 residues; it did not isolate any bind-ing peptides from the RPLs. Taken together, these resultsindicate that AFLs are suitable for identifying linear epitopes,and perhaps discontinuous ones.

This example indicates that AFLs are a better choice formapping linear epitopes, but draws no conclusions aboutwhich library type is better for mapping discontinuousepitopes. It is not surprising that a discontinuous-epitopebinding MAb did not isolate binding phage from the RPLsused by Fack et al. (59), as neither of the libraries they usedcontained disulfide constraints. It has been our experiencethat this type of Ab always selects constrained peptides.Other examples that use AFLs for epitope mapping includethe work of Bentley et al. (50), who mapped several immuno-dominant antigenic regions on the outer capsid protein ofAfrican horsesickness virus using chicken and horse PCAbs,and the work of Holzem et al. (60), who identified a seven-residue epitope for a MAb against tobacco mosaic virus.

In a unique approach, Huang et al. (61) used AFLs tomap discontinuous epitopes on porcine rotavirus. Theyproduced AFLs expressed from the VP7 gene; cDNA wasself-ligated (presumably into multimers), digested with DNa-seI, and then blunt-end ligated to limiting amounts of phagevector. This produced clones containing short stretches ofVP7 sequence as well as clones containing discontinuousVP7 sequences ligated together, which might function as dis-continuous epitopes. The library was screened with sevenvirus-neutralizing MAbs thought to bind discontinuous VP7epitopes; however, no binding clones were isolated. Thelibrary was also screened with PCAbs from mice, rabbitsand convalescing pigs, and sequences of the isolated clones


mapped to portions of VP7. The PCAbs selected clonescontaining mostly single fragments as well as some in-framefragments that were flanked by out-of-frame ones; only twoout of the 23 analyzed clones bore in-frame fragments fromtwo regions of the protein. One composite clone, isolated bymouse PCAbs, contained fragments from two antigenicregions of VP7; however, there is no evidence to suggest thata single Ab bound to both regions. It appears that this librarydid not produce mimics of discontinuous epitopes, as none ofseven discontinuous-epitope binding MAbs selected anyclones; this probably occurred because only a small proportionof this library expressed the desired composite sequencesin-frame. Perhaps an approach like this would succeed withadditional work. For instance, a DNA shuffling may improvethe mixing of and variation within fragments (62), and inaddition, ligated fragments could be selected for ORFs (21)before cloning DNA fragments into phage display vectors.

IV.B. RPLs for Epitope Mapping

RPLs have been used to identify ligands for anti-protein Absagainst linear and discontinuous epitopes. As explainedabove, peptides that cross-react with linear epitopes are mostdependably obtained, since they usually do not require a largenumber of CBRs to bind Ab. Linear epitopes are also easier tomap, once the CBRs are known. Discontinuous epitopes aremore difficult to identify, since CBRs on the sequencemay relate to different parts of the protein, and many CBRsmay be needed to structure a peptide ligand (see below). Thus,peptides that cross-react with discontinuous epitopes aremore difficult to obtain and may require several stages ofdevelopment.

Work from our laboratory with human MAb b12 illus-trates this point. MAb b12 neutralizes a broad range ofHIV-1 isolates by binding to the envelope protein gp120 at alarge discontinuous epitope, which overlaps with the CD4-binding site (63). Screening of a gp120-based AFL with MAbb12 yielded no binding peptides (Parren, personal communi-cation, 2003). Bonnycastle et al. (53) used MAb b12 to screen


a panel of 11 RPLs, and two clones were identified that sharea 5-residue sequence. Based on this consensus, Zwick et al.(29) produced and screened two sub-libraries, and the best-binding peptide (B2.1) shared a significant amount ofsequence homology with the D loop of gp120. Later, Ala repla-cement studies showed that most of the homologous residueswere CBRs (Ollman-Saphire E, Montero M, Menendez A, vanHouten NE, Irving MB, Zwick MB, Parren PWHI, Burton DR,Scott JK, Wilson IA, manuscript in preparation). In support ofthis homology with the gp120 D loop, the crystal structure ofMAb b12 bound to the peptide showed that, on a gross struc-tural level, the peptide binds MAb b12 at the same paratopesubsite as that deduced for the D loop (63). Surprisingly, fiveof the seven CBRs in the peptide did not contact the Ab, andthus, were required for the correct structure of the peptide(Ollman-Saphire E, Montero M, Menendez A, van HoutenNE, Irving MB, Zwick MB, Parren PWHI, Burton DR, ScottJK, Wilson IA, manuscript in preparation). Of the two CBRsthat directly contacted the Ab, only one matched with a CBRon the D loop, and replacement of this Asp residue with Alaablates MAb b12 binding to gp120. Thus, except for a commonAsp residue, the B2.1 peptide and the D loop of gp120 do notappear to share a commonmechanism for binding to MAb b12.

In an ambitious study, Bresson et al. (57) identifiedCBRs derived from screening phage libraries and alignedthem to regions on the auto-Agn, thyroperoxidase (TPO), tolocalize regions that contribute to an immunodominant, dis-continuous epitope. Four phage libraries were screened witha MAb that binds a discontinuous epitope on TPO and com-petes for binding with most anti-TPO sera. Three consensusmotifs were identified, and Ala replacement was used todetermine the CBRs of these peptides. Consensus sequenceswere aligned to TPO, identifying five regions that potentiallycontribute to MAb binding. In many cases, the alignment wasquestionable, since spaces had to be added to the CBRs fromthe peptides to match ‘‘homologous’’ residues on TPO. Foreach of the five regions identified, four to ten residues werereplaced with the amino acid sequence from the correspond-ing region of a homologous protein. This was meant to change


the sequences in these regions without affecting the overallfolding of TPO. These TPO mutants were tested by westernblot for binding to three MAbs and a rabbit PCAb, and byELISA for binding to sera from patients with the autoimmunethyroid disorders, Grave’s disease andHashimoto’s thyroiditis.Patient sera showed reduced or no binding to four of the TPOmutants, and the MAbs did not bind to two of these mutants.The authors concluded from these results that four regionscontribute to the immunodominant epitope on TPO.

It is unclear whether the loss in binding was due tomutations in the immunodominant region of TPO or to dena-turation, since the authors did not determine whether theamino acid replacements affected global folding of TPO. Otherstudies, using homolog replacement (see Ref. 49), showeddecreased MAb binding to certain homolog replacementmutants. When MAbs were tested for binding to single Alareplacement mutants in these regions, there was no effecton the degree of MAb binding, and this indicated that homo-log scans may introduce some disruptive effects to proteinfolding. The study by Bresson et al. (57) presents an interest-ing method for identification of immunodominant epitopes.However, further studies could strengthen the conclusions;for example, the putative immunodominant epitope could besubjected to Ala replacement scans.

Several laboratories have used MAb-binding motifs iden-tified from RPLs to map a linear epitope and determine itssecondary protein structure. Stern et al. (64) isolated twoMAbs (GV1A8 and GV4D3) from a mouse immunized withHIV-1 gp120. These were shown to bind to the N-terminalportion of gp120 between residues 1 and 142 by probingprotease-digested, reduced gp120 on a western blot with bothMAbs. To map the MAb-binding site on gp120, a 20-mer RPLwas constructed and screened with MAb GV1A8. A singlephage clone (j35) was identified, but the sequence did notmatch gp120, and therefore, shared amino acids could notbe identified. To obtain more clones, the authors re-screenedthe library under less stringent conditions, and they isolatedan additional 18 clones that shared a common di-peptidemotif (Leu=Ile-Trp). Alignment of these clones to the


N-terminal region of the gp120 sequence, based on thedi-peptide motif, revealed other residues that were sharedwith gp120, but were not shared among all the phage clones;this identified the general region of the MAb GV1A8 epitope.Each of the 19 clones was analyzed based on its homologywith gp120, and clone f66 exhibited the highest homology.The gp120 epitope encompassed residues 108–113, asrevealed by alignment with clone f66, and confirmed byMAb binding to overlapping peptides that covered this region.However, even though clone f66 shared the greatest homol-ogy with the gp120 epitope, it exhibited lower affinity forthe MAb GV1A8 than did five other binding clones. This ledthe authors to continue to look for determinants of homologywith gp120.

The alignment of all 19 phage clones with gp120 revealeda pattern of conserved residues at four positions along thegp120 sequence (HxxIxxLW). Only one clone (f35) bore thefull sequence motif, and this clone exhibited the highest affi-nity for MAb GV1A8. The analysis indicated that the fullbinding motif (HxxIxxLW) was consistent with two turns ofan alpha-helix, and it was hypothesized that the C1-regionof HIV-1 had an alpha-helical secondary structure. Acomputer model of residues 89–117 of gp120, configured asan alpha-helix, placed the CBRs for MAb GV1A8 on a contig-uous surface, whereas this was not the case when this regionon gp120 was modeled as a beta strand. To further confirmthe analysis, it was postulated that the second MAb(GV4D3), which binds to gp120 in the same region as MAbGV1A8, would also select a series of phage clones withsequences consistent with an alpha-helical motif. This wasconfirmed by screening the same 20-mer RPL with MAbGV4D3. Seven phage clones were isolated and analyzed byalignment with gp120. The authors concluded that this regionalso bore a helical motif. However, fewer clones were isolatedfor this MAb, and the peptides had overall lower homologyscores than those isolated for MAb GV1A8. The authors didnot confirm their structural hypothesis with physical ana-lyses, such as circular dichroism. However, it may not be sur-prising that a putative alpha-helical motif was identified,


given that helicity is a common structure among proteins. Thecomputationally derived hypothesis, that residues 93–112 ofHIV-1 gp120 form an alpha-helix, was later confirmed bythe crystallographic structure of gp120 (65).

In a similar vein, work by Ferrieres et al. (66) used RPLsto show that residues in a linear epitope that are not CBRscan contribute to Ab affinity, probably by conferring struc-tural stability to the peptides. The authors first used Alareplacement to identify the consensus sequence, YXTEPH,for a linear epitope on troponin I; replacement of residuesflanking this region did not appear to affect MAb binding.They next screened an RPL with the MAb, yielding peptidesbearing the YXTEPH consensus; however synthetic versionsof the phage-displayed peptides were found to have lower affi-nities than synthetic peptides bearing the native epitope. Toimprove affinity of the phage-selected sequences, the authorsreplaced sequences flanking the consensus on a phage-selected sequence with those flanking the consensus on nativetroponin I. Although Ala replacement of these sequences didnot affect MAb binding, transfer of either the N-terminal orthe C-terminal region from troponin I to flank the consensussequence of a phage-selected peptide improved affinity forimmobilized MAb. In some cases, improved affinity correlatedwith improved structural stability, as shown by circulardichroism studies. Thus, longer sequences, which apparentlydo not contain CBRs, can affect binding affinity. As with thestudy of Stern et al. (64), the results of Ferrieres et al. (66)indicate a global, multiresidue effect that is consistent withstructural stabilization, which in some cases could bedetected.

Our laboratory has produced further evidence ofsequences that confer increased affinity due to global effectsthat do not depend on discrete CBRs. Working with a MAbthat recognizes a linear epitope on the membrane proximalregion of the HIV-1 envelope protein gp41, Menendez et al.(68) constructed and screened RP sub-libraries that extendedthe peptide on either side of the core epitope (DKW). Two Alaresidues were placed N-terminal to the DKW and Ser wasplaced C-terminal to DKW in the sub-libraries so that the


high-affinity epitope, ELDKWA, could not be produced; thiswould make selection of higher-affinity peptides dependentupon the N- and C-terminal sequences from the sub-library.Thus two libraries, X12AADKWS and AADKWSX12, were con-structed and screened with the MAb under stringent condi-tions that would select the tightest-binding clones. No suchclones were selected from the X12AADKWS library, butthree relatively tight-binding clones were selected from theAADKWSX12 library. To perform in-solution affinity studiesusing a BIAcore instument, the sequences of the displayed pep-tides were transferred to the N-terminus of the maltose-bind-ing protein of E. coli, which allowed monovalent display inthe context of a fusion protein (67). The authors showed thatthe affinity of the peptides could be increased to levels similarto the affinity of gp41, by replacing the two Ala residues preced-ing the DKWmotif with Asp and Leu; this supports the conclu-sion that there is overlap between the sites on theMAb that thepeptides and native epitope bind. The affinities were twoorders of magnitude greater than that of a synthetic peptidebearing the ELDKWAS epitope flanked by Gly residues.

Ala replacement studies on the phage-displayed versionof two of the sequences showed little effect of singly changingany of the last four residues. Yet, deletion of the last three resi-dues of the same peptides decreased binding of the Fab versionof the MAb significantly, by >50% in one case and > 80% inthe other. (As Fab binds phage-borne peptides monovalently,its apparent affinity is less affected by peptide polyvalencythan IgG, which is bivalent. Thus it is more sensitive to differ-ences in affinity.) These studies now await crystallographicstudy to determine the role of these C-terminal sequences inenhancing the affinity of binding to the MAb. Fusion of thepeptides to smaller proteins should allow structural studies(NMR, circular dichroism) that may reveal the degree to whichstructural stabilization plays a role in Ab affinity.

Thus, thework of Stern et al., Ferreries et al. andMenendezet al. (64,66,68) has elucidated three different approaches toanalyzing the role of non-CBRs in promoting Ab affinity.There remains much to be done in revealing the connectionbetween affinity and structural stability in all of these cases.


In a recent study, Enshell-Seijffers et al. (69) describea novel computational approach to mapping discontinuousepitopes on the surfaces of proteins with known tertiarystructures, using the sequences of MAb-binding peptides iso-lated from RPLs. The algorithm is based upon two generalassumptions. First, that Ab-selected peptides contain aminoacid pairs that occur at a higher frequency than the aminoacid pairs in a RPL. Second, that those selected amino acidpairs will occur in close proximity in the cognate protein epi-tope. The analysis was applied to three MAbs that recognizedoverlapping epitopes on the gp120 envelope protein of HIV-1.The epitope of one of the MAbs (17b) was previously shown tocomprise four beta-strands of the HIV-1 envelope proteingp120, based on a cocrystal structure of MAb 17b bound togp120 (65). A constrained 12-mer RPL was screened withMAb 17b, and 11 peptides, having no homology to gp120, wereselected and shown to bind by dot blot. The frequency of eachamino acid pair in the peptide sequences was calculated. Themost frequently occurring pairs were assumed to representeither continuous residues on the cognate epitope or non-continuous amino acids that are brought together by proteinfolding. The authors identified all amino acid pairs that arewithin a short distance of each other on the Agn surface,and matched the amino acid pairs identified from the RPLscreening to them. Twenty-six pairs from the RPL screeningmapped into a region that overlapped with the crystallogra-phically defined MAb 17b epitope. The region comprised fourdiscontinuous regions, containing seven of the eight knownsegments that were observed to contact the MAb, some ofwhich spanned up to eight contact residues. Another MAb(CG10), which competes with MAb 17b, was thought to bindan epitope on gp120 that overlaps with the MAb 17b epitope.Twenty-eight peptides were selected from the library screen-ing and analyzed as described above. The identified epitopecontained five residues known to affect MAb CG10 bindingand, as suspected, was located near the MAb 17b epitope.

A discontinuous epitope-mimic peptide was constructedby fusing the four gp120 regions identified in the analysis intoa single, 25-residue peptide that was displayed on phage. This


reconstructed epitope bound to MAb CG10 and competed withgp120. The authors did not determine its affinity, the regionsinvolved in binding, nor the CBRs within them. Ala replace-ment studies would further clarify the mechanism of binding,as would a crystal structure of MAb CG10 bound to the peptidemimic; the latter would also show the degree to which thepeptide mimics the cognate epitope on gp120. This approachshows promise as an alternative to homolog replacement stu-dies for identifying discontinuous epitopes for MAbs that bindproteins with known structures. Moreover, it may serve as aviable approach to the structure-based design of epitope-mimic peptides for vaccine and diagnostic applications.

As alluded to above, the main promise of epitopemapping is its application in the design of vaccines that elicitthe production of Abs against specific epitopes. There are anumber of reports of peptides that, when used as immuno-gens, will elicit Abs that cross-react with a cognate Agn, pre-sumably at a single, targeted epitope. Typically, thesequences of these peptides are based on a cognate linear epi-tope. AFLs are a superior source of immunogenic mimics ofprotein epitopes. Being derived from the cognate Agn, AFsare more likely to elicit Abs that will cross-react with cognateAgn. However, as most Abs recognize discontinuous epitopes,Ab-binding peptides may not be present in a given AFL. Alter-nate approaches may be required if AFs are not found for aspecific Ab. For example, Enshell-Seijffers et al. (69) (seeabove) used sequence data from clones selected from anRPL to identify the structure of a discontinuous epitope,and then designed a peptide mimic of that epitope, based onthat structure. This construct was never tested as animmunogenic mimic, but may prove to be one.

Another approach to epitope-targeted HIV-1 vaccines isbeing explored by Pantophlet et al. (70). Their goal is to pro-duce a vaccine that will elicit Abs having the properties ofMAb b12, a human MAb that neutralizes a broad range ofHIV-1 primary isolates by binding to the CD4-binding siteand blocking the virus–receptor interaction. Instead of usinga small domain to mimic a discontinuous epitope on gp120, thisgroup engineered gp120 to block Ab binding to unrelated,


immunodominant epitopes and to bind strongly to MAb b12,but not to the non-neutralizing MAb b6, which also binds theCD4-binding site. Immunodominant epitopes on gp120 wereblocked by the addition of glycosylation sites, and the CD4-binding site was engineered to decrease binding to MAb b6and to bind more tightly to MAb b12. This engineered gp120thus does not bind CD4, the MAb b6 or a number of othergp120-binding Abs, but binds well to MAb b12, and immuniza-tion studies with it are underway. The engineering of gp120was based on work by Ollman Saphire et al. (63), who produceda computationally derived model of the MAb b12 structuredocked onto the gp120 structure of Kwong et al. (65), as wellas extensive gp120 mutagenesis studies that differentiatedresidues on gp120 that are important for MAb b12 bindingfrom those that are important for MAb b6 binding (71).

Both the approaches of Pantophlet et al. (71) and Enshell-Seijffers et al. (69) are structure-based epitope-targeted appro-aches, designed to create immunogens that elicit Abs against aspecific epitope. It may also be possible to target the productionof a specific Ab (as opposed to Abs against a specific epitope), byusing a prime-boost approach (see below). After priming withcognate Agn, production of the desired Ab is selected and ampli-fied during the boost with an Ab-specific peptide. For thisapproach to work, the desired Abs must be elicited during thepriming immunization. One caveat is that complex Abs, suchas those having extensive somatic mutation and=or those thathave rare specificities, may be difficult to target by any of theapproaches described. The next section will briefly introducethe topic of vaccines, and follow it with an examination of theuse of peptides, derived from AFLs and RPLs, as immunogensfor targeting the production of Abs of predetermined specificity.

V. PHAGE DISPLAY LIBRARIES FOR VACCINEDEVELOPMENT

Since Jenner and Pasteur, vaccines have evolved from poorlydefined ‘‘bacterial soup’’ mixtures into molecularly well-defined formulae (72). Some vaccines comprise attenuated


or whole killed organisms (e.g., cholera and pertussis) andothers ‘‘detoxified’’ exotoxin (e.g., tetanus toxoid, TT). Morerecently, vaccines have evolved into purer formulae compris-ing partially purified subunits (e.g., acellular pertussis andinfluenza virus vaccines), recombinant whole Agn (e.g., hepa-titis B Agn) and polysaccharide conjugate vaccines (e.g.,pneumococcal capsular polysaccharide conjugated to TT)(72,73). All vaccines today elicit neutralizing Abs as a partof their protective action; some, like the anti-toxin vaccines,rely completely on neutralizing Abs for protection. Molecularvaccines target the Ab response against a specific Agnmolecule (i.e., induce the formation of Agn-specific Abs), andtheoretically, could target the production of Abs against a spe-cific epitope. As discussed in the section above, epitope map-ping can provide leads for targeting the production ofspecific Abs against an epitope during immunization. In thefollowing section, we review the application of epitope map-ping to vaccine design. Specifically, we explore how differenttypes of phage libraries may be applied to find vaccine leadsthat target epitopes and their corresponding Abs.

V.A. Phage Libraries for Vaccine Design

The primary application of phage libraries to vaccine designis in identifying ligands that target biologically active Absand=or the epitopes they recognize. An Ab-targeted vaccineelicits Abs that have the functional properties of a bio-logically active Ab (i.e., neutralization), whereas an epitope-targeted vaccine elicits Abs against a specific, biologicallyrelevant epitope. Both approaches are viable, and perhaps,often reflect different aspects of the same process. Peptideshave great potential for targeting both specific epitopes andAbs.

The idea of Ab-targeted vaccines evolved from the idioty-pic network theory developed by Jerne and others. Jerne (74)defined an Id as ‘‘a set of epitopes displayed by the variableregions of a set of antibody molecules.’’ Abs differ from oneanother by virtue of their variable region sequences. Theseform the Ab combining site, which to some degree overlaps


with the Ab paratope (the regions of the Ab that contact theAgn’s epitope). The Ab combining site also overlaps with theId (the regions of the Ab that are immunologically distinctand can elicit an Ab response in syngenic animals). On a func-tional level, the Id is serologically defined by the Abs producedagainst it, called anti-Id Abs. Some anti-Id Abs can be thoughtof as carrying a ‘‘mirror-image’’ of the Ab’s paratope, and thusmimicking the corresponding epitope on the cognate Agn;these Abs will compete with Agn for binding to the Id Ab.Other anti-Id Abs may bind to regions of the Id that areimmunologically distinct from the paratope, and will notcompete with Agn for binding to the Id Ab.

By immunizing with a single Id Ab (i.e., a MAb, or highlyrestricted PCAb response), it is theorized that at least someanti-Id Abs will mimic the Agn, in binding the paratope. Byimmunizing with anti-Id Abs that can compete with Agn,some of the resulting anti–anti-Id Abs may behave as the ori-ginal Id Ab, and bind to Agn as well as the original Id. Suchanti–anti-Id Abs are thought to be similar to the original IdAb (75,76). Thus, several groups have sought to use anti-IdAbs to target the production of specific Abs. This has beenespecially attractive as an alternative way of producing Absagainst T-cell independent, weakly immunogenic Agns likepolysaccharide (77) and more recently DNA (78), as well asself-Agn targets for cancer vaccines (79,80). For reviews, seeRefs. 81, 82.

Until very recently, there had been little effort to useanti-Id Abs to target production of anti-protein Abs. In animpressive piece of work, Goldbaum et al. (75) demonstratedthat immunization with a single Id Ab, followed by immuniza-tion with a PC anti-Id preparation, can target the productionof specific anti–anti-Id Abs against a discontinuous proteinepitope. Rabbits were hyper-immunized with Id MAb D1.3,which binds a discontinuous epitope on hen eggwhitelysozyme (HEL). The resulting anti-Id IgGs were affinity pur-ified on MAb D1.3 and extensively absorbed with mouse IgGto remove any anti-isotypic or anti-allotypic Abs. Mice wereimmunized with these rabbit anti-D1.3 Id PCAbs and theresulting anti–anti-Id sera were tested for binding to HEL


and to the D1.3-anti-Id MAb, E5.2, which is known to structu-rally mimic the D1.3 epitope on HEL (83). The anti–anti-Idsera bound to HEL, but bound better to MAb E5.2. Hybrido-mas were produced from immune mice, and two anti–anti-Id MAbs (AF14 and AF52) were isolated that bind bothHEL and MAb E5.2. Interestingly, the amino acid sequencesof MAbs AF14 and AF52 differed from D1.3 by 10 residues atmost, indicating they were probably derived from the samegerm line genes and shared the same type of V generearrangements (75). These results indicate that it is possibleto target the production of specific Abs using anti-Id Absas structural mimics of a cognate discontinuous epitope.However, there are problems associated with using anti-IdAbs in vaccines. The affinity of anti–anti-Id MAbs, AF14and AF52, was greater for the anti-Id MAb E5.2 than forHEL. This was especially true for MAb AF14, whose associa-tion constant (Ka) for MAb E5.2 was three orders of magni-tude greater than that for HEL. Also, immunizing withwhole Abs can be expensive, and as Abs are highly conservedproteins, not all individuals will mount a strong responseagainst an Ab-based vaccine. Finally, there is the chance thatsuch a vaccine would elicit an autoimmune response.

Despite these drawbacks, targeting a specific Ab is adesirable way to approach vaccine design, for some purposes.Because of tolerance, the immunogenic regions on an anti-IdAb are limited to its Ab combining site; thus the Ab responseis limited to a relatively small region on a very large immuno-gen. This is reminiscent of the work of Pantophlet et al. (70),who blocked the immunodominant sites on gp120, limiting itto a single epitope; perhaps epitopes as restricted as thiswould act as anti-Id Abs and elicit a highly restricted Abresponse, with MAb b12-like reactivity being a major compo-nent of it. As with the approach of Enshell-Seijffers et al. (69),peptides and AFs represent an alternative to large moleculeswith restricted immunogenicity as components forAb-targeted vaccines. In this case, a peptide or AF is more likea hapten, in being small and having only a few epitopes. Ascompared to a larger, more complex Agn, a hapten-likepeptide or AFwould elicit only a limited number of reactivities.


The following sections discuss how the polypeptides and pep-tides identified from AFLs and RPLs can be used to targetthe production of Abs having specific functions.

The success of an Ab-targeted vaccine relies on choosingan Ab that has a desired biological activity. Biological activityhas many forms; for example, in the case of HIV-1, some Absblock infection by neutralizing the virus, in vitro, or protectagainst infection, in vivo. An in vitro-neutralizing Ab mayblock infection by binding viral receptors. However, in vivoprotection is mediated by a number of factors. For example,Abs may opsonize the surface of a bacterium, thus triggeringcomplement mediated lysis, or phagocytosis by macrophages.Abs such as these would be useful to target with a vaccine.Second, the targeted Ab must be producible in the first place;if an Ab has extensive somatic mutation and=or uncommonvariable-gene usage, then this approach may not work.

Ab- and epitope-targeted vaccination is especially suitedto situations in which immunization with whole Agn producesa negative or blocking effect. For example, immunization withcognate Agn may induce Abs that interfere with a neutraliz-ing Ab response (32) (see below) or, in the case of anti-cancervaccines, some MAbs may stimulate cell division when theybind a surface receptor whereas others may block it (55).Ab- and epitope-targeted vaccine approaches can avoid theproduction of such unwanted Abs and focus the immunesystem to produce effective specificities.

Most current vaccines are designed to prevent infectionby certain bacteria or viruses; however, Ab- and epitope-tar-geted vaccines may be used against diseases whose origin isnot infectious. For example, some vaccines are being designedto treat cancer (55) or to block autoimmune diseases (84).Other studies are aimed at producing tolerance against speci-fic allergens, thus preventing allergic reactions (85). Newwork also suggests that vaccines can prevent degenerativediseases like Alzheimer’s disease (86). The methods discussedin the following section can be applied to these diverseailments as well as to infectious diseases. We discuss howAFLs and RPLs can be used in Ab- and epitope-targetedvaccine design.


V.B. AFLs and Epitope-Targeted Vaccines

AFs, selected by specific Abs, are effective immunogens thatcan be used to target the production of Abs against specificepitopes. AFLs yield fragments that are more likely to elicitAbs that cross-react with cognate Agn than peptides foundin RPLs. This was demonstrated by Matthews et al. (3) usingT4 phage as a model pathogen in a murine system. Mice wereimmunized six times with T4 phage, and their purified IgGswere affinity purified on T4 phage. These IgGs were usedto screen 12 phage libraries comprising a set of 11 RPLsand one T4 genome AFL. Of the 35 unique clones isolatedfrom the RPLs, only two unique peptides contained 5-mersequences that mapped directly to T4 proteins. Consensussequences that did not map to T4 proteins were observed infour sets of clones derived from different libraries. Screeningof the AFL with the anti-T4 IgG resulted in 16 unique clonesranging from 15 to 97 residues in length, and they were allin-frame fragments of T4 ORFs.

Thirteen phage-displayed peptides from RPLs and eightfrom AFLs were used to immunize mice. Sera were testedafter three immunizations for both anti-peptide and anti-T4 Abs. Cross-reactivity with T4 was defined by the percentdrop in Ab titer after serum Abs were adsorbed with wholeT4 compared to mock-adsorbed sera. Since intact T4 phageparticles were used, Abs against internal T4 proteins maynot have been removed from sera, if they were present. Serafrom only two RP-immunized groups cross-reacted with T4at a 10-fold dilution; this level of cross-reactivity was notconsidered significant. Only four AFs elicited anti-peptidetiters, and of these, two elicited 40–100% cross-reactivitywith T4 phage. The low anti-peptide Ab response elicitedby AFs was likely due to their low copy number on phage.To increase anti-T4 cross-reactive Ab titers, six AFs (includ-ing the two that elicited cross-reactivity) and one RP werefused to the N terminus of an internal T4 protein, as a car-rier for immunization. Three AFs elicited high-titer Absagainst T4 phage, and the percent cross-reactivity elicitedby the RP was lower than those elicited by the AFs. This


study demonstrated that, although peptides derived fromRPLs are immunogenic and elicit anti-peptide Abs, theyare less likely than AFs to elicit Abs that cross-react withthe cognate Agn. This approach should be valuable in defin-ing epitopes for subunit vaccines. AFLs should be consideredfor identifying vaccine leads prior to using RPLs as a sourceof ligands for specific Abs.

MAbs against the self-oncoprotein ErbB2 can eitherstimulate or inhibit tumor growth depending on the epitopethey bind. Yip et al. (55) used an ErbB2 AFL to identify ananti-cancer vaccine-lead peptide. Seven anti-ErbB2 MAbswere used to select AFs from two ErbB2 libraries, one ofwhich was constrained by flanking Cys residues; the AFs ran-ged from 16 to 50 residues in length. Four of the MAbs did notselect phage from the AFLs; most likely, they bound to discon-tinuous epitopes not present in the AFLs. Three murine MAbs(N12, N28, and L87) isolated AFs, and these MAbs weretested for their ability to effect the growth of several breastcancer cell lines. MAb N12 inhibited growth of the four breastcancer cell lines tested, whereas MAb N28 inhibited growth ofonly two and enhanced the growth of the other two. Thus,MAbs N12 and N28 affected cell growth, presumably, bybinding ErbB2.

MAb N12 probably bound to a discontinuous epitope,since it repeatedly selected AFs comprising the same 55-resi-due sequence corresponding to residues 531–586 of ErbB2,and it did not bind to any overlapping 15-mer synthetic pep-tides that covered this region. The AFs selected by MAbsN28 and L87 shared a 20-mer overlapping epitope region;however, their alignment was not shown. MAb L87 alsobound to one peptide from a panel of 15-mer overlapping syn-thetic peptides, and MAb N28 did not, indicating that theseMAbs bind overlapping, but non-identical epitopes. This isconsistent with MAb N28 having biological activity, but notMAb L87. A competition ELISA between these two MAbswould further confirm whether they share identical or over-lapping epitopes, but this was not mentioned. Three RPLswere screened to identify CBRs for the MAbs but no peptideswere selected.


Mice were immunized five times with GST fusions toeither the 55-residue AF selected by MAb N12, or to a 20-resi-due AF selected by MAbs L87 and N28. Sera were analyzed byELISA after three and five immunizations for anti-AF Absand anti-ErbB2 Abs. Only the 55-residue AF elicited Abs thatcross-reacted with ErbB2, even though the Ab response wassix- to sevenfold lower than that against the AF. Cross-reactive, protein A affinity-purified IgGs were tested for theirability to inhibit growth of one of the breast cancer tumor cellline, BT474. IgGs from the 55-residue AF-immunized miceinhibited 85% of cell growth compared to 65% inhibition bythe positive control MAb 4D5, 15% inhibition by negative con-trol anti-GST IgGs, and 45% inhibition by MAb N12, the Abused to select the AF. The approach discussed here targetsthe production of Abs against a folded protein subdomain,as opposed to a single epitope or the production of one specificAb. However, these results illustrate that using AFs asvaccine leads may avoid the production of Abs against unde-sirable epitopes. In addition, there are instances in whichMAbs do not select fragments from AFLs. RPLs may provideligands for these MAbs, as discussed in the next section.

V.C. RPLs and Ab-Targeted Vaccines

AFsare a good source ofAb-specific ligands.However, not all Abswill selectAFs, especially those that binddiscontinuous epitopes,whose CBRs are distant on the primary protein structure. Inthese situations, RPLs may be a better a source of Ab-specificligands. Peptides derived from RPLs interact with an Ab para-tope by mechanisms that are not necessarily identical to themechanismofbinding to thecognateepitope,but theycanbeveryspecific to that Ab; their specificity can make them a good choicefor targeting particular Abs. Although peptides from RPLs maybe a good alternative to anti-Id vaccines, in targeting specificAbs against a discontinuous epitope, peptides derived fromRPLsmay cover only a small portion of the paratope in comparison toan anti-Id Ab which can cover the entire paratope. Buildingsub-libraries to extend the coverage of the paratope by peptideligand is one way to overcome this problem (30–32).


There are several examples that demonstrate thatimmunization with peptides derived from RPLs results in pro-tection against infection. Yu et al. (56) demonstrated protec-tion from infection with an encephalopathogenic strain ofmurine hepatitis virus (E-MHV) after immunization withpeptides isolated from RPLs. Three MAbs (5B170, 5B19,and 7–10A) that neutralized E-MHV in vitro and protectedmice in vivo against intracerebral challenge with E-MHVwere used to screen a panel of 13 RPLs. MAbs 5B170 and5B19 are known to bind to linear epitopes on the E-MHVS2-glycoprotein subunit, and MAb7-10A is predicted to binda discontinuous epitope on the same antigen. MAb 5B170selected five unique clones and a six-residue consensus motifwas identified. MAb 5B19 selected one peptide that containedthe same six-residue motif as the peptides selected by MAb5B170. This was assumed to be the consensus sequence forMAb 5B19 even though only one binding sequence was iden-tified. An ELISA, in which phage clones were titrated ontoplate-bound MAb, showed that both MAbs 5B170 and 5B19bound to the same phage clones. The S2-glycoprotein wasexamined for homology to the consensus sequence isolatedby MAb 5B170 and MAb 5B19, and a high degree of similaritywas found for several different MHV strains, indicating thatthese are conserved residues on the virus. Thus MAbs 5B19and 5B170 compete for binding to E-MHV and probably sharea conserved, linear epitope.

The putative discontinuous epitope-binding MAb 7–10Aisolated 10 unique phage clones from three libraries. A con-sensus sequence for MAb 7–10A was not easily identified,since some peptides contained only two residues and othershad common residues spaced apart from one another. Onepeptide, which did not share consensus to other peptidesshared six out of nine residues with the S2-glycoprotein.Phage bearing these peptides were tested for binding toMAb 710-A in the presence of reducing agent; disulfide reduc-tion ablated binding to three peptides, consistent with bind-ing to a discontinuous epitope.

C57BL=6 and BALB=c mice were immunized four timeswith 10 selected phage-displayed peptides (five for MAb


7–10A, one for MAb 5B19 and four for MAb 5B170) at 14-dayintervals. Mice were given an intra-cerebral viral challengewith 10LD50 MHV-A59, which induces 100% mortality afterfive days. Only three of six C57BL=6 mice immunized withclone 9.1 (selected by MAb 5B170) survived the viral chal-lenge. BALB=c mice immunized with clone 9.1 and all otherphage-immunized groups died before 10 days. Serum analysisby western blotting revealed that only mice immunized withclone 9.1 produced serum Abs that bound to recombinantS2-glycoprotein this was also true for BALB=c mice eventhough they were not protected from infection. To determinethe mechanism of protection, immunized mouse sera wereused in neutralizing, antibody-dependent cell-mediated cyto-lysis and antibody-dependent complement-mediated lysisassays. No activity was detected in any of these assays at50-fold dilutions. Thus, this study showed that only onephage-borne peptide, which mimicked a linear epitope on E-MHV, elicited partial protection from viral challenge,although the mechanism of action was not clear. The protec-tive response was also strain dependent.

Working with sera from HIV-1-infected people, Scalaet al. (51) theorized that specific Abs protected long-terminfected (LTI) patients from progression to AIDS, and thatpeptides that bound to these Abs could be used as a vaccinelead to target their production. They used serum IgGs fromdisease-free LTI patients to isolate peptides from two RPLs.Five peptides that bound with high frequency to LTI patientsera and low frequency to AIDS patient sera were chosenfor further study; the peptides also bound sera from SHIV-infected macaques. The peptide sequences were comparedto the envelope proteins, gp120 and gp41. One clone(p217) was mapped to the gp120 C2 region whereas another,(p197) mapped to the cluster I epitope of gp41. The p195 pep-tide was questionably mapped to the V1 region of gp120. Theremaining two peptides (p287 and p335) did not bear homol-ogy to gp120 or gp41. Serum Abs from a LTI individual wereaffinity purified with each of the phage clones, and were thenused in western blots to identify whether they bound to gp160and=or gp120. Abs purified on all of the clones bound both


gp120 and gp160, with the exception of clone p197, whichbound to neither. Purified phage clones were used to immu-nize mice, and the purified immune IgGs were shown to neu-tralize three isolates of HIV-1 in vitro (two were lab-adaptedstrains, and one was a primary HIV-1 isolate). It is puzzlingthat immune Ab against a cluster I epitope mimic (p197)would neutralize infection by an HIV-1 primary isolate, sincestrong Ab responses are typically present in HIV-1 infectedpeople at all levels of progression (LTIs, rapid, and normalprogressors), and are not associated with viral neutraliza-tion. Nevertheless, this work clearly demonstrates theidentification of peptides that are frequently recognized bythe sera of HIV-1-infected people, and produced preliminarydata indicating that they could be used to produce Abs thatcross-react with HIV-1 envelope, and perhaps, neutralizethe virus.

In an extension of the work described above, Chen et al.(87) tested the five phage clones identified by Scala et al. (51)for their ability to protect rhesus macaques from infectionwith SHIV89.6PD, a pathogenic SIV–HIV chimera. Fivemacaques were immunized five times, at 10-week intervals,with pools of the five phage-displayed peptides, and fournegative-control macaques were immunized with WT phage.Ab titers against synthetic versions of each phage-displayedpeptide in the experimental group ranged from 800 to24,000, and anti-gp120 Ab titers ranged from 100 to severalthousand. One phage-peptide-immunized monkey producedlow anti-phage-peptide Ab titers that did not cross-react withgp120. At 44 weeks, naive, WT-phage-immunized and pep-tide-phage-immunized macaques were challenged withSHIV-89.6PD. All groups became infected by the virus, asdetermined by a peak in viremia at 12 days, which coincidedwith a drop in CD4þ T-cells counts. However, four of the pep-tide-immunized macaques showed a significant increase inCD4þ T-cells and lower virus levels compared to naive orWT-phage-immunized controls. Analysis of the postchallengeAb response showed that five of the peptide-immunized maca-ques elicited an anti-gp140–89.6 Ab response with titers inthe 100,000 range. In contrast to the macaques immunized


with the phage-peptides, one macaque in each of the controlgroups (naive and WT-phage immunized animals) developedan anti-gp120–89.6 Ab response with significantly lowertiters (100–3000). All of the naive and WT-phage-immunizedmacaques developed severe AIDS-like illness six weeks post-challenge. Among the phage-peptide-immunized macaques,one developed AIDS-like symptoms that progressed at similarrates to control macaques, this was the same macaque thatdid not show any prechallenge Ab response to gp120. Anothermacaque, in the peptide-immunized group, became ill withacute glomerulonephritis, but disease pathology was unre-lated to SHIV infection; this animal most likely died froman unrelated pre-existing condition. The three remainingpeptide-immunized macaques were alive and healthy at 270days postchallenge (87). Thus, immunization with a mixtureof phage displaying five different peptides elicited Abresponses that bound synthetic peptides representing the fivephage-displayed peptides, cross-reacted with gp140, andappeared to reduce the disease caused by infection with SHIV89.6PD.

The examples discussed above demonstrate that peptidesfrom RPLs have potential as vaccine leads, but require signif-icant optimization before a functional vaccine could be devel-oped. For example, the peptides used by Chen et al. (87)elicited immune responses that prevented development ofan AIDS-like disease in SHIV-infected macaques, but wereunable to prevent infection. Furthermore, only half of themice immunized with a peptide isolated by Yu et al. (56) wereprotected from MHV infection. It may be that these peptideswould more effectively target the production of the desiredAbs if used in an alternative immunization approach, suchas the prime-boost strategy outlined in Sec. VI.B.

V.D. Peptide Ligands for Anti-CHO Abs andTheir Use in Anti-CHO Vaccines

Vaccines against CHOs present a very different set ofproblems from those directed against proteins, and some ofthese may be addressed with peptides from RPLs. The best


defense against bacteria, such as Haemophilus Spp. andStreptococcus pneumoniae, is to produce Abs that will opsonizetheir polysaccharide coat. Unfortunately, polysaccharides areweakly immunogenic, comprising repeating units having lim-ited epitope diversity. Thus, anti-CHO Ab responses are oftenweak in infants, the elderly and immuno-compromised indivi-duals. Polysaccharides are T-independent Agns, and thereforeform weak to no immunological memory in these age groups(73). Several recently released vaccines comprise polysacchar-ide that is chemically conjugated to the immunogenic carrierprotein, TT. The immune response against such conjugatesis similar to the hapten-carrier response (described inSec.VI.A) in that it elicits a T-cell-dependent immune responseagainst TT, and creates immunological memory in both Agnbinding B- and T-cells. Since B-cells recognize both the TT car-rier and the CHO conjugated to it, B-cell memory is formedagainst both Agns. Thus, stronger and longer-lasting Abresponses are elicited in response to multiple vaccinations,due to the T-cell-dependent secondary response. Notwith-standing these advances, there continue to be problems withthe development of some CHO vaccines. First, it is not guaran-teed that the anti-polysaccharide Ab within this population isincreasing proportionally to the anti-TT Abs, and sometimesanti-CHO responses may remain weak. Second, it can bedifficult to impossible to synthesize complex CHO Agns, andsometimes native CHO Agns are difficult to isolate in a pureform. Third, some viable CHO targets, such as oligosaccharidemoieties on LPS, are difficult to separate from larger toxicmolecules. Finally, in some cases, only some of the Abs elicitedagainst a CHO Agn will confer the desired biological activity.This can be due to the production of Abs having the wrongisotype (e.g., non-opsonizing Abs), or to Abs with the wrongspecificity (i.e., they do not neutralize the targeted pathogen).

A number of alternative strategies have been explored todevelop vaccines with strong and protective anti-CHOresponses (reviewed by Lesinski and Westerink (88)). Theearliest strategy explored was use of anti-Id Abs as vaccinesthat target CHO Agns. Immunization with peptides fromRPLs represents a more recent twist to this approach, since


it removes the difficulties associated with developing, produ-cing and immunizing with Id Abs, while targeting the IdAb. Several studies have used peptides for CHO vaccines, asthey may be optimized for desired characteristics thatenhance structural and=or functional mimicry of the cognateCHO epitope (e.g., see Ref. 32).

The nature of CHO-mimicry by peptides that bindanti-CHO Abs has been explored in the structural studies ofVyas et al. (10) using the MAb SYA=J6, which is specific forthe LPS O-Agn polysaccharide of Shigella flexneri Y. Theydetermined the crystallographic structures of Fab SYA=J6bound to either an eight-residue peptide (8) or to a syntheticoligosaccharide analog of the cognate CHO epitope (9). Thestructures were quite similar on a gross structural level, witheach Agn occupying the same site in the Ab combining site.The peptide made six hydrogen bonds and 126 van der Waalscontacts with the Fab, whereas the pentasaccharide madeeight hydrogen bonds and 74 van der Waals contacts. For bothligands, 57% of the contacts were made with the light chain.However, for the peptide, most of these contacts were withthe CDR3 loop, which forms a shallow hydrophobic cavity;this interaction is not shared by the pentasaccharide.Furthermore, only half of the 37 contacts that are sharedbetween the peptide and pentasaccharide are similar in type(i.e., polar–polar, nonpolar–nonpolar, polar–nonpolar). NineFab residues made contact solely with the peptide comparedto two solely with the pentasaccharide. The peptide also inter-acted with a greater number of water molecules, 14 comparedto 2 for the pentasaccharide (10). The authors concluded thatthe peptide had better shape-complementarity for the Fabthan the pentasaccharide (as reflected by the larger numberof contacts with the Fab), and that the peptide is thus not astructural mimic of the pentasaccharide. These differencesin structural interaction may explain our preliminary immu-nizations with the peptide (as a synthetic peptide conjugateand as a phage-displayed peptide), which do not elicit Absthat cross-react with the LPS (our unpublished data). It ispossible that by using this peptide in a prime-boost immuni-zation strategy, as described below, Abs will be elicited that


behave like SYA=J6, in cross-reacting with the bacterial LPSand the peptide.

The first example of the immunogenic mimicry of a CHOAgn by a phage-displayed peptide was reported by Phaliponet al. (89). They used two protective IgA MAbs that bind toShigella serotype 5a LPS, to screen two RPLs (the peptidesin one of the libraries were constrained by flanking Cys resi-dues). MAb C5 selected 13 peptides and MAb I3 selected six.Five of the peptides selected by MAb I3 also reacted with MAbC5. This type of cross-reactivity, between MAbs that bind thesame CHO, has been observed by other groups (90) and mayreflect a restricted Ab response against the CHO. BALB=cmice were immunized six times with one of the 19 phage-displayed peptides, and sera were tested for binding to LPSby western blot. Only two peptides elicited cross-reactivitywith the LPS: p100c (selected by MAb I3) and p115 (whichwas selected by MAb C5 and cross-reacted with MAb I3).The cross-reactivity was relatively weak, as the anti-LPStiters were measured at 100-fold dilutions, whereas theanti-phage titers were measured at 10,000-fold dilutions.The serum Abs are serotype specific, since Abs bound toLPS from Shigella serotype 5a bacteria but not to serotype2a bacteria (89). No in vitro neutralization or in vivo chal-lenge studies were performed to analyze the protective prop-erties of the Abs elicited by the peptides. Nevertheless, theauthors clearly showed that immunizations with phage-displayed peptides can elicit Abs that specifically cross-reactwith an LPS.

Immunization with whole Cryptococcus neoformansglucuronoxylomannan (GXM), a component of the capsularpolysaccharide, elicits both protective and non-protectiveAbs. Non-protective Abs can block the function of protectiveAbs; moreover, the Abs produced by immunization with theGXM do not protect against infection. However, some anti-GXM MAbs do protect against infection. Beenhouwer et al.(32) used an Ab-targeted, prime-boost immunization strategyto elicit the production of protective Abs against GXM. Protec-tive MAb 2H1 was used to screen phage libraries for ligands,and peptide PA1 was identified. It did not elicit anti-GXM Abs


when used as an immunogen on its own. To improve thebinding of MAb 2H1 to PA1, a sub-library was built aroundthe PA1 core residues by adding one random residue at thecenter of the sequence and six random residues at theC- and N-termini. Six rounds of screening isolated peptideP206.1, which had an estimated Kd of 4 nM compared to300nM for the original PA1 peptide. Direct and competitionELISAs showed that P206.1 also bound to other anti-GXMprotective MAbs, whereas it did not bind to a non-protectiveMAb. The P206.1 peptide was used to make a number of dif-ferent immunogens. It was produced as a synthetic peptideand conjugated to TT (P206.1-TT), as a multiple antigenicpeptide (P206.1-MAP) in which multiple copies of the peptidewere attached to a poly-lysine backbone, and as a free peptide.Immunizations with P206.1-TT, P206.1 alone, and P206.1-MAP did not elicit anti-GXM titers above those seen in con-trol mice immunized with TT alone.

It was hypothesized that P206.1 could specifically stimu-late B-cells producing protective Abs from a pre-existingpopulation of B-cells producing Abs against both protectiveand non-protective epitopes. Thus, mice were first immunizedwith a low dose of GXM conjugated to TT (GXM-TT). This, intheory, activated B-cells that produce Abs against the protec-tive epitopes, as well as B clones whose Abs recognize non-protective epitopes. No anti-GXM Abs were detected afterthe prime. To amplify the production of Abs against onlythe protective epitopes, mice were boosted with eitherP206.1-TT, P206.1 alone, P206.1-MAP or TT alone, andanti-GXM titers were determined after 14 days. Only thegroup that was primed with GXM-TT and boosted withP206.1-TT developed anti-GXM titers that were significantlyhigher than the TT boosted control group. These anti-GXMserum Abs cross-reacted with both GXM and peptide, asshown by competition ELISA, and did not bind to de-O-acety-lated GXM, which is preferentially bound by non-protectiveAbs. Immune mice were not challenged with C. neoformansto assess protection. This elegant work clearly shows thatprime-boost immunizations can significantly augment theproduction of peptide-targeted Abs.


A study by Pincus et al. (90) demonstrated immunogenicmimicry by a peptide selected by a protective MAb (S9)against the type-III capsular polysaccharide (type-III CPS)of group B Streptococcus (GBS). MAb S9 isolated two peptidesfrom an RPL, and the sera of mice infected with GBS bound toa synthetic version of one of the peptides. This peptide wasconjugated to ovalbumin (OVA), bovine serum albumin(BSA) and keyhole limpet hemocyanin (KLH), and the conju-gates were used to immunize mice in complete Freund’s adju-vant (CFA). Immune sera, diluted 1000-fold, appeared tocross-react with whole GBS and with the type-III CPS, com-pared to preimmune sera. However, the analysis was incon-clusive, since titers were not calculated, and control serafrom carrier-only immunizations were not analyzed. Suchcontrols are important, as glutaraldehyde-conjugated KLHand unconjugated KLH have been shown to elicit non-specificCHO-binding Abs (91). Thus, the carrier proteins may haveproduced false-positive reactivity; data from carrier-onlyimmune sera would clarify this. In vitro or in vivo challengestudies were not performed to confirm biological activity ofthe immune sera. Recent NMR studies have analyzed thestructure of the peptide in solution (92), and ultimately, thesemay help to reveal the structure of the corresponding epitopeon the type-III CPS.

The most conclusive study to date showing protective Abproduction by an immunogenic-mimic peptide is described inRef. 93. They produced hybridomas specific for meningococcalgroup A polysaccharide (MGAPS), and MAb 9C10 was used toscreen an RPL. After three rounds of selection, the sequencesof 60 phage clones were analyzed and a dominant sequencerepresenting 38 of the clones was identified. Direct ELISAshowed that human hyper-immune sera against group CMeningococcus bound to phage displaying the peptide almostas well as to MGAPS, and the sera did not bind to controlphage. A synthetic analog of the peptide was incorporatedinto proteasomes prepared from synthetic lipopeptides andouter membrane complex vesicles from group B meningococci.The first immunizations comprised a dose response studywhere mice were immunized three times with either 1, 5,


10, 50 mg peptide=proteosome complex, with proteosome aloneor with 5-mg MGAPS. A 5-mg dose of the peptide=proteosomecomplex was shown, by titration ELISA, to give the highestanti-MGAPS Ab titers. These Ab titers were much higherthan those for the group immunized with MGAPS alone.The second immunization regimen comprised a prime-booststrategy, in which mice were primed with 1, 5, 10, or 50-mgpeptide–proteasome complex and then boosted at day 28 with5-mg MGAPS. This part of the study did not include anMGAPS control. Although titers were calculated in two differ-ent ways (the dose response study used 50% of maximal titersand the prime-boost study used 25% of maximal titers), theprime-boost approach for the 5-mg dose group appeared to eli-cit similar to higher anti-MGAPS Ab titers than did threeimmunizations with the peptide–proteasome complex aloneor with MGAPS alone. Serial dilutions of immunized mousesera were tested for in vitro bactericidal activity in the pre-sence of complement, with data reported as the dilution ofserum required to kill 50% of the bacteria. Sera from groupsimmunized twice with the peptide–proteasome complex killed50% of bacteria at a dilution of 25, compared to a dilution of 23for sera produced one week after a single MGAPS immuniza-tion. Mice that were primed with peptide–proteasome com-plex and boosted with MGAPs killed 50% of bacteria at asignificantly greater dilution of 150.

Thus, several studies have shown that RPLs are a viablesource of peptides that can serve as both functional mimics,and in some cases, immunogenic mimics of CHOs; however,the structural mechanisms behind this cross-reactivity areunclear. Several studies have demonstrated that peptideimmunization can elicit Abs that cross-react with cognateCHO (89,90,93). Others were only successful when a CHO-prime and peptide-boost were used (32). Despite these suc-cesses, there is little evidence to suggest that cross-reactiveAbs protect in vivo. Grothaus et al. (93) performed in vitrostudies confirming the anti-biotic capabilities of the peptide-elicited Abs. The Ab response to CHOs is sometimes restrictedand may contribute to the success of peptide immunogensfor anti-CHO, Ab-targeted vaccines. Perhaps this restriction


enabled prime-boost immunizations, such as the one conductedbyBeenhouweret al. (32), to succeed. It remains to be seen if theprime-boost approach will work for specific anti-proteinMAbs.

In general, vaccines made from whole-protein immuno-gens are preferable to subunit vaccines. However, if the wholeAgn is difficult to produce, or if a non-immunodominant epi-tope is to be targeted, then AFLs may be a useful source ofimmunogens that will target specific epitopes on the Agn.There are instances in which an AF cannot be isolated, espe-cially for some MAbs directed against discontinuous epitopes.In these circumstances, and for CHO-binding MAbs, RPLsmay yield Ab targeting peptides. Very likely, epitope- andAb-targeting AFs and peptides from RPLs require modifica-tion to become strongly immunogenic. In the next section,we discuss considerations for immunization strategies withpeptides derived from these libraries, their formulation asvaccine components, and finally, the analysis of immuneresponses to vaccination with them.

VI. DEVELOPING IMMUNOGENS FROMPEPTIDE LEADS

The studies described in the above sections focus on theisolation and analysis of peptide ligands for MAbs and PCAbsthat bind protein and carbohydrate Agns. In the majority ofthese examples, peptides isolated from AFLs and RPLs wereused for mapping epitopes to proteins or for immunizationwith the purpose of eliciting Abs that cross-react with the cog-nate Agn. The desired outcome of this approach is to developvaccines that elicit biologically relevant Abs when used asimmunogens. One caveat of peptides as immunogens is that,by themselves, they are rarely immunogenic. The followingsection reviews methods for incorporating vaccine-lead pep-tides into effective immunogens. Since this chapter focusesmostly on targeting Abs, we will discuss methods for enhan-cing the humoral response, rather than cellular immunity,even though both aspects of the immune response should beconsidered when designing a vaccine.


The process of formulating an Ab-binding peptide into avaccine involves numerous steps. The first step, after isolat-ing the peptide, is to test the immunogenicity of the peptide.The peptide must be incorporated into a carrier for immuniza-tion and other components of the vaccine must be decided on.These include dose, adjuvant, immunization route, andtiming between immunizations. Each of these componentsmay require optimization to enhance anti-peptide responseand cross-reactivitywith the cognateAgn. Step two is thoroughanalysis of the immunized sera. This involves measuringanti-peptide, anti-cognate Agn, and anti-carrier Ab titers bydirect ELISA. Cross-reactive Abs that bind both peptideand cognate Agn are determined by direct ELISA and shouldbe confirmed by competition ELISA. Results from theseassays indicate the requirement for further optimization ofthe vaccination strategy or specific vaccine components.

Cross-reactivity with cognate Agn by peptide-elicited Absis the first indication that a vaccination strategy is producinga desired outcome. Cross-reactivity is only significant if thepeptide-elicited Abs have the appropriate biological effect.The later stage of vaccine-lead development requires strin-gent testing of the biological activity of the Abs elicited bythe peptide vaccine. In vitro neutralization assays followedby in vivo challenge studies confirm that Abs being targetedby a peptide vaccine share the same biological activityas those Abs elicited after immunization with the vaccine.These criteria must be met if a vaccine-lead peptide is to beconsidered for further development into a vaccine.

VI.A. Carrier Proteins and the Inductionof T-Cell Help

A humoral immunogen requires both B-cell epitopes (BCE),and TH-cell epitopes (TCEs) to trigger Ab responses thatinclude affinity maturation and immunological memory. Thepeptides we have discussed above are functional mimics ofBCEs, given that they bind the Ab paratope. These peptidesmay serve as a replacement for a cognate BCE in a vaccinewhose purpose is to target the production of specific Abs;


however, additional components are required to make BCEpeptides immunogenic. Carrier molecules are proteins thatcan serve as sources of TCEs, and when attached to small-molecule BCEs, like haptens and peptides, can make themimmunogenic. Traditionally, a hapten or peptide is chemicallycoupled to a carrier protein, and immunization with the con-jugate produces an Ab response against the carrier and themolecules coupled to it. More recently, carriers can be arecombinant protein (including a phage) to which a BCE pep-tide is fused. Choices for carrier proteins include traditionalimmunogens such as OVA, BSA, TT, and KLH. The mosteffective carrier proteins are those that contain many TCEs,as they will elicit Ab responses in a variety of species andstrains. However, as mentioned, carrier proteins also containBCEs that elicit the production of Abs. These may interferewith analysis of immune sera. For example, May et al. (91)detected Abs that cross-reacted with GXM from C. neofor-mans in sera from mice immunized with KLH. These Abs pre-ferentially bound to KLH over GXM in competition ELISAsand, when used to passively immunize mice, did not protectthem from challenge with C. neoformans.

Self-proteins are typically non-immunogenic; however,they can be made immunogenic by the addition of exogenousor foreign TCEs. For example, the highly conserved proteinubiquitin fused to an exogenous TCE was able to elicit anAb response specifically against a V3-based peptide fromHIV-1 (which served as the targeted BCE) (94). ExogenousTCE coupled to self-protein is critical to this approach, asthe former drives helper T-cell responses, whereas B-cellclones that would recognize the self-protein are probablydeleted. Thus, carriers comprising an immunogenic TCEand a self-protein avoid the production of Abs, and allowthe Ab response to be focused against exogenous B-cellepitopes.

Vaccine carriers are not exclusively limited to proteins,so long as they incorporate immunogenic TCEs. Other meth-ods for vaccine delivery include liposomes, comprising singleor multilamellar bilayer membrane vesicles where Agn ismembrane bound or within the intermembrane spaces. The


immunomodulation effects of liposomes depend on the compo-sition of the micelle and the types of proteins and adjuvantsincorporated into the lipid bilayer. They are particulate, andthey target APCs and create Agn depots which allow longerexposure of the immune system to the Agn (95). Liposomes(called virosomes) are approved for human use in influenzavaccines (96).

MAPs can also serve as carriers, if they include TCEpeptides in their structure. MAPs comprise multiple syntheticpeptides, of single or multiple specificity, which are coupled toa poly-lysine backbone (97). Like self-proteins, they overcomeproblems associated with immunodominant BCEs on carrierproteins, and can be suitable for focusing Ab responsesagainst a BCE peptide. Microparticles, made of biodegradablepolymers, may also serve as carriers; these are suitable forsingle dose administration since they create long-term Agndepots and do not require repeated boosts. Synthetic peptideleads can be spray dried onto the microspheres (98).

The topic of vaccine carriers covers a large field, whichwill not be discussed further. Since this chapter focuses onphage-displayed peptides for vaccine development, we turnto the advantages and disadvantages of filamentous phageas carriers for BCE peptides and AFs.

The use of filamentous phage as carriers for peptides iswell established. In 1988, de la Cruz et al. (99) first demon-strated the use of phage for eliciting Ab responses against adisplayed peptide. They immunized rabbits with phage dis-playing a peptide from Plasmodium falciparum fused to pIIIand discovered that they were immunogenic, even in animalsimmunized without adjuvant. Later, Meola et al. (100) com-pared immune response to several peptides derived fromscreening RPLs with anti-hepatitis B surface Agn (HbsAgn)MAbs. Peptides were fused to pVIII or pIII, made into MAPs,and conjugated to Hepatitis B virus core peptide (HBV core)and to human ferritin. Phage immunogens were administeredwithout adjuvant, whereas adjuvant was used in immuniza-tions with the HBV core and ferritin conjugates and theMAPs. It was found that pVIII-displayed peptides consis-tently elicited higher anti-HbsAgn Ab titers than the other


carrier proteins, with pIII display coming in close second. Allof the recombinant carriers elicited higher Ab titers againstrecombinant HbsAgn than the synthetic peptide MAP.

Bastien et al. (101) demonstrated that a protectiveimmune response could be induced by phage displaying a pep-tide, derived from respiratory syncytial virus (RSV), and wasknown to protect from infection after immunization. Micewere immunized with phage displaying the RSV peptide onpIII. They were then challenged with RSV, and their lungswere checked for the presence of virus five days postchal-lenge. All of the phage-immunized mice had cleared the virusfrom their lungs, whereas controls had not. These earliersuccesses have led many groups to use phage as carriers forvaccine-lead peptides derived from phage libraries (Sec. V).

More recently, Grabowska et al. (102) used MAb H5,against herpes simplex virus (HSV) glycoprotein G, to isolatethree peptides from an RPL. Consensus motifs deduced fromthe three peptides aligned to a linear epitope on the native gly-coprotein G amino acid sequence. Prior to immunization,phage clones were either treated to remove LPS or leftuntreated. BALB=c mice were immunized with various dosesof the pooled phage clones. After two immunizations, serumAb responses against both phage and glycoprotein G weremeasured in ELISA by comparing OD490 of sera at 1000-folddilution. Ab titers were not determined. Groups of mice immu-nized with untreated phage clones produced anti-phage andanti-glycoprotein G Ab responses in a dose-dependent man-ner, whereas mice immunized with treated phage had muchlower Ab responses. This did not affect the protective effectof the Abs in vivo, since mice immunized with high doses(100 and 70 mg) of either treated or untreated phage were com-pletely protected from lethal challenge with HSV-2. Three ofthe groups immunized with lower doses of phage (50 and 10mg) had only a 30% survival rate. The exception was the fourthlow-dose group, which received 50 mg of treated phage, andhad a 65% survival rate. The level of serum Ab did not corre-late with survival rates in this study, indicating that aspectsof the immune response other than serumAb were responsiblefor protection. Although LPS did not increase the ability of


mice to survive viral challenge, its presence did enhance bothanti-phage and anti-glycoprotein-G Ab responses. LPS is aB-cell mitogen, and confers this advantage to the immuno-genic effects of phage as a carrier for BCE peptides. Furtheradvantages are discussed in the section below.

VI.A.1. Advantages of Using Phage as a Carrier

Although filamentous phage are unlikely to be used in com-mercial vaccine formulations, there are several advantagesto using them as carriers for BCE peptides. Phage are easilyamplified to large quantities, they can be rapidly purified byPEG precipitation, and they can be further purified by CsCldensity-gradient centrifugation. This makes the productionof phage immunogens a rapid, inexpensive approach to test-ing peptide immunogenicity, compared to synthesizing pep-tides and chemically conjugating them to carrier proteins.Filamentous phage particles have the general dimensions ofa virus but are non-infectious to animal tissues; this makesthem good pathogen mimics. Also, as the phage surface com-prises mostly the outer 10 residues of the major coat protein,pVIII, phage coats are homogeneous and restricted to a few B-cell epitopes (103). This can result in lower titers against thephage carrier compared to more complex carriers, such asOVA (van Houten NE, Zwick MB, Menendez A, Scott JK,manuscript in preparation).

As mentioned above, carrier proteins enhance theimmune response by providing TCEs. There is evidence thatphage bearing exogenous TCEs can target either the cellularor humoral branches of the immune system, or both. Williset al. (104) reported that the humoral response to phage-displayed peptides is helper T-cell dependent. They immu-nized nude, heterozygous, and BALB=c mice with phage dis-playing a peptide, and tested by ELISA whether classswitching occurred in anti-peptideAbs (i.e., whether IgGswereproduced). Nude mice, lacking T-cells necessary for B-cellclass switching, produced only IgM, whereas BALB=c miceproduced IgG, and the heterozygous mice produced a mixtureof IgM and IgG. They also demonstrated that T-cells wererecruited by WT phage in the presence or absence of adjuvant.


Furthermore, phage can be genetically manipulated toenhance T-cell help and recruit cytoxic T-cell responses. DeBerardinis et al. (105) engineered phage to simultaneously dis-play a helper T-cell epitope (p23) and a cytotoxic T-cell (CTL)epitope (RT2) derived fromHIV-1 reverse transcriptase. Phagedisplaying both epitopes elicited cytotoxic activity in humanT-cell lines (as determined by 51Cr release), whereas phagedisplaying the RT2 CTL epitope alone did not induce cytotoxicactivity. For other examples, see Refs. 106 and 107.

Phage are effective carrier proteins for syntheticpeptides. Our laboratory engineered an additional lysine resi-due near the N-terminus of pVIII to be used as an accessibleprimary amine for amine-reactive cross-linking agents(Zwick, unpublished data). This allowed the conjugation of apeptide to every two to three pVIII molecules on the phage,and this corresponds to 1000–2000 peptides per phage. Afterthree subcutaneous (SC) immunizations, the anti-peptide Abresponse exceeded that of the anti-phage Ab response bytwofold (van Houten NE, Zwick MB, Menendez A, Scott JK,manuscript in preparation).

VI.A.2. Disadvantages of Using Phage as aCarrier Protein

The copy number of a displayed peptide on the phage coat is amajor determinant controlling the Ab response against agiven peptide. Peptides displayed by pIII (via Type 3 vectors)have copy number fixed at three to five copies per phage, andproduce relatively weak anti-peptide Ab responses. Type 8vectors produce phage displaying 2500–3000 copies of peptideper phage, and in theory, this should make them excellentfor peptide-targeted immunization. However, pVIII cannottolerate more than four or five additional residues withoutblocking viral assembly (28,108,109). This severely limitsthe type of peptide that can be used for immunization withType 8 vectors. The Type 88 vector, f88.4, produces hybridphage having a displayed peptide copy number varying from1% to 15% of the total pVIII, with other such vectors (110)producing peptide copy numbers reaching 30% (personalcommunication, Cesareni). Variation in the copy number of


a peptide displayed in a Type 88 system depends on the aminoacid composition and length of the peptide, and is probablyrelated to the efficiency with which peptide–pVIII fusionsassemble, along with WT pVIII, onto the virion as it emergesfrom the inner membrane of the bacterium. Thus, longer frag-ments, such as those found in AFLs, are displayed at lowerfrequencies and, although immunogenic, may not induce adetectible Ab response due to low copy number (3). Peptide–pVIII fusions that assemble poorly onto phage result in vir-ions bearing peptide at low copy numbers, making them poorimmunogens compared to phage that display peptides in highcopy number.

Analysis of the copy number of peptide–pVIII fusions isunreliable; SDS-PAGE and amino acid analysis has been usedfor this purpose, but each has its shortcomings. Probably thebest method is separation of peptide–pVIII fusions from WTpVIII by SDS-PAGE (111). The goal is to separate the twotypes of protein, and to determine the relative concentrationof protein in each band. This is done by running side-by-sidedilutions of the phage and identifying peptide-fusion and WTpVIII bands in different dilutions that have comparable inten-sities. The percentage of peptide–pVIII fusion is then calcu-lated using the dilution factor between these two samples.However, in some instances, the recombinant and WT formsof pVIII will not resolve from each other, making thisapproach impossible. Amino acid analysis can be used to cal-culate peptide copy number, based on the fact that pVIIIforms most of the protein mass of the virion. As pVIII is 50residues long, and has a restricted amino acid composition(e.g., Cys, His, and Arg are absent), peptide copy numbercan be calculated from comparing the molar amounts of twocategories of amino acids in the analysis: (i) the amount ofan amino acid that is present in the peptide, but absent frompVIII, or the amount of an amino acid that is present in pVIIIand absent from the peptide, and (ii) the amount of an aminoacid that is present in both pVIII and the peptide.

Given a peptide of interest that is displayed at low copynumber, in our hands, it has been very difficult to improve dis-play density on phage. We have attempted this for several


clones by using different vectors and different promoters withonly limited success (unpublished data). Thus, it is likely thatlow peptide copy number cannot be raised on a routine basis.This makes the use of synthetic peptides attractive; however,this approach comes with a different set of problems. As men-tionedabove, in our experience, theaffinity of apeptide selectedfrom an RPL is typically much lower as a synthetic peptide,than for its recombinant-fusion counterpart. This problemhas occurred with several synthetic peptides that we havederived from RPLs. Most likely, these peptides adopt a morestable conformation in the context of a fusion protein, whereastheir free-peptide counterparts lack persistent structure (112).

Several studies demonstrate that phage may not be anideal carrier protein in all circumstances and several optionsshould be investigated to find the ideal immunogen. Forexample, Yip et al. (113) tested the immunogenicity of a55-residue peptide that binds to an anti-ErbB2 MAb. Theauthors compared peptide fusions to the N-termini of pIIIand pVIII, and to the C-terminus of GST, which producedthree to five copies per Type 3 phage, an unknown copy num-ber per Type 88 phage, and two copies per GST dimer. TheGST-peptide fusion elicited the highest anti-peptide andanti-ErbB2 Ab titers. The low titers produced by the phageprobably reflected low copy number from both vectors: pIIIdisplay is low, and the low titers elicited by the Type 88 vectorare likely due to the length of the peptide. Rubinchik andChow (114) also compared different carrier proteins expres-sing residues 47–64 from the Staphylococcal superAgn, toxicshock syndrome toxin-1 (TSST-1). They compared fusions toGST, the outer membrane porin protein (OprF) of E. coli, pIIIand its synthetic counterpart conjugated to BSA (a relativelyweak carrier protein). All carriers, with the exception of pIII,induced high Ab titers against TSST-1; however, only Absfrom the GST immunized group inhibited binding of the syn-thetic peptide to MHC class II and inhibited TSST-1-inducedT-cell proliferation.

As mentioned above, Willis et al. (104) showed thatphage elicit T-cell responses. However, our unpublishedobservations have lead us to question the strength of the


T-cell response against phage, as compared to commonly usedcarrier proteins, since this was not done in the earlier study.Thus, we immunized Balb=c mice with a synthetic peptideconjugated to phage or to the carrier protein, OVA. The Abresponse against the peptide was far stronger in the groupimmunized with the peptide-OVA conjugate. Moreover, Abresponse against the phage carrier plateaued after five immu-nizations, whereas the response against the OVA carrier wasstill on the increase after seven immunizations. From this, wededuced that phage elicit a weaker helper T-cell responsethan OVA. Thus, phage can produce good Ab responsesagainst a displayed foreign peptide or AF, but have the draw-backs that the copy number of the displayed peptide or AF canbe low, and that they may not elicit very strong T-cellresponses, as compared to more traditional carrier proteins.

VI.B. Vaccine Formulation

Choosing a carrier molecule is the first step in formulating avaccine for a lead peptide or AF. This is followed by developingan immunization strategy. A complete vaccine comprises adju-vants, timing of boosts, boosting immunogens, dose, and injec-tion route. The section belowgives a brief overviewof how thesedifferent components are incorporated into a vaccine.

VI.B.1. Immunization Strategy

The immunization strategy should be one of the first consid-erations in vaccine formulation. ‘‘Straight’’ immunization, inwhich a single Agn is used for several immunizations, is com-monly used. This elicits an immune response focused onimmunodominant regions of the immunogen. This approachworks well if a peptide or AF immunogen elicits Abs thatcross-react with cognate Agn, or if immunodominant Absare being targeted. However, certain peptides do not elicitAbs that cross-react with cognate Agn after straight immuni-zation. To elicit Abs that bind to such peptides and cross-reactwith cognate Agn, a prime-boost approach could be considered(Fig. 1).


Figure 1 Strategy for targeting the production of specificAbs usinga prime-boost immunization approach. Prime: The first immuniza-tion comprises cognate-Agn conjugated to a TCE-containing carrierprotein (e.g., TT) or to a foreign TCE peptide. This elicits a PCresponse from B-cells (B) that bind to numerous epitopes on the cog-nate Agn, as well as, a small proportion of B-cells that produce the Abthat is being targeted (TAb). (1) The foreign TCE, presented byB-cells (both TAb and B, only TAb is shown), activate a TCE-specificpopulation of helper T-cells (TH) that signal to B, Tab, and other TH

cells to differentiate andmature. (2) Memory TH-cells (THM), specificfor the foreign TCE, are elicited during the prime immunization andawait subsequent encounterswith the sameTCE. (3) BothB andTAbdifferentiate into memory B-cells (BM and TAbM, respectively) thatmay be selectively amplified by the boosting immunogen. Plasmacells that secrete Ab for B and TAb are also produced (not shown).Boost: The second immunization comprises a TAb-specific peptideimmunogen. It may be expressed as a recombinant fusion protein orchemically conjugated to the same carrier protein used in the prime.(1) The boosting immunogen contains the same foreign TCE as thepriming immunogen and stimulates ThM cells elicited during theprime. The peptide portion of the immunogen amplifies TAb-specificmemory B-cells (TAbM) from the pre-existing population of memoryB-cells. (2) The stimulated TAbM cells differentiate into Ab-produ-cing plasma cells (TAbP), increasing the specific Ab titer in serum.(3) The pre-existing population of BM cells that produce the non-targeted Ab will not be amplified since there is no protein epitopefor them to bind to in the boosting immunogen.


A prime-boost immunization strategy is a method forenhancing the production of targeted Abs that recognize aspecific epitope or region on a selected Agn. This approachcomprises a priming immunization with the cognate Agnfollowed by a boosting immunization with a peptide or AFthat specifically binds the targeted Abs. Thus, the primingimmunization activates and expands a B-cell population thatincludes those that produce the targeted Abs. Also in theprime, helper T-cells provide signals that drive the differen-tiation of naive B-cells, which recognize the cognate Agn, intomemory B-cells and plasma cells. The boosting immunogen,comprising an Ab-specific peptide or epitope-specific AFcoupled to a carrier protein, amplifies only the memory B-cellsthat produce the targeted-Ab. To produce a strong response,the boost should be driven by memory T-cells that wereformed after the prime; this requires that the priming andboosting immunogens carry a strong, identical TCE. Such astrategy has been used by Beenhouwer et al. (32) to elicitthe production of specific Abs against the capsular polysac-charide of Cryptococcus neoformans. Animals were primedwith capsular polysaccharide coupled to TT, and then boostedwith an Ab-specific peptide coupled to TT to produce thetargeted Ab.

VI.B.2. Adjuvants

Adjuvants enhance the immune response and can selectivelyactivate specific branches of the immune system (i.e., humoralvs. cellular). Cox and Coulter (95) have classified adjuvantsinto two broad categories: particulate and non-particulateadjuvants. Others have categorized adjuvants by their sourceof origin (96). Particulate adjuvants create Agn depots thatstimulate Agn presenting cells (APCs) and provide short tolong-term exposure of the immunogen to the immune system.They include: aluminum salts (e.g., alum), water-in-oil emul-sions (e.g., Freund’s complete and incomplete adjuvants; CFAand IFA, respectively), liposomes and biodegradable micro-particles. Particulate adjuvants can be further modified bythe addition of non-particulate adjuvants that have specific


immunostimulatory effects. For example, IFA is modified tobecome CFA by the addition of heat-killed Mycobacteriumtuberculosis, which acts as an irritant and enhances the activ-ity of macrophages (1). Other particulate adjuvants, suchas alum, contribute an innate immunostimulatory effect.Alum stimulates TH2 responses resulting in increased IgGproduction (96).

Non-particulate adjuvants have a direct effect on thecellular components of the immune system. These includecytokines, block copolymers, and bacterial components suchas monophosphoryl lipid A (MPL) and CpG oligodeoxynucleo-tides (95,96). Cytokines target specific immune cells; forexample, IL-12 activates TH1-dependent cell-mediated immu-nity, granulocyte-macrophage colony stimulating factor(GM-CSF) activates dendritic cells, and IL-4 stimulatesB-cells (115). Block copolymers such as those included inTiterMax2 target APCs (95).

Only aluminum salts (e.g., alum), virosomes, and MF59are for use in humans, but many others adjuvants are avail-able for research (96). Adjuvants commonly used in researchinclude CFA and IFA (see above), alum (see above), TiterMax(see above), quil A (a saponin-based adjuvant that inducesboth humoral and cellular responses), and Ribi2 (which con-tains MPL and induces a strong TH1 response) (95,116–118).Certain adjuvants, like CFA, are commonly used but areassociated with severe side-effects, including granulomaswhich can ulcerate into abscesses (116). While CFA andIFA can elicit responses against weak immunogens, cautionand restraint should be used because of their severe side-effects. Furthermore, Kenney et al. (118) discovered thatAbs elicited by CFA bound to denatured epitopes, whereasAbs induced by other adjuvants were more selective for thefolded protein Agn.

The most appropriate adjuvant for any given applicationshould be empirically determined. Bennett et al. (116) com-pared TiterMax, alum, Ribi, and CFA and found Ab titersinduced by TiterMax were comparable to those of CFA, butresulted in fewer side-effects. Kenney et al. (118) comparedquil A, Ribi, CFA, alum, and SAF-1 and discovered quil A and


Ribi produced a more suitable immune response for makinghybridoma, even though CFA produced higher Ab titers. Inour hands, alum has induced higher Ab titers in Balb=c miceto the model Agn HEL, compared to CFA or TiterMax(unpublished data).

VI.B.3. Immunization Route, Dose, Timing,and Number of Boosts

Since pathogens use different entry routes, so must protectivevaccines. Immunization route should be considered so that theprotective response is focused in areasmost likely to encounterpathogen. Vaccines elicit systemic immune responses by sub-cutaneous (SC), intramuscular (IM), or intraperitoneal (IP)delivery, whereas vaginal, rectal, oral, or nasal routes confermucosal protection. Phage have been successfully used in mul-tiple routes including IP, SC, oral (PO) or intra-nasal (IN)(119,120). In our experience with phage immunizations inmice, we have had the greatest success with SC immunizationsas compared to IP for eliciting systemic Ab titer; this may bedue to more effective access to dendritic cells (DCs). Severalimmunization routes may be tried to optimize immuneresponse and adjuvants should be chosen according to route;for example, cholera toxin is often used to aid the formationof a mucosal immune response (reviewed in Ref. 121).

Other considerations for vaccine development includedose, timing of boost, and the number of boosts. These typi-cally need to be empirically determined and are Agn depen-dent; however, there are some general points that may beconsidered. For example, a lower dose of Agn will selectivelyactivate B-cells producing high-affinity Ab, compared to ahigher dose that activates a greater breadth of response. Thisis an important consideration when targeting specific Abs,since a discrete immune response may be preferred. Timingof the boost may be critical if a prime-boost immunizationstrategy is being considered. One method of assaying the suc-cess of a prime-boost immunization strategy is to measure Ablevels against the cognate Agn before and after the boostingimmunization to determine if there has been an increase in


titer. This can only be done if the immune response againstthe cognate Agn has peaked and then dropped. If the boostoccurs too soon after the prime, then an increase in Absagainst the cognate Agn may be due to the natural kineticsof the priming immunization, rather than the boosting immu-nogen. Furthermore, several boosting immunizations may berequired to amplify the targeted Abs.

VI.C. Assays

The primary goal of a vaccine is to elicit a protective immuneresponse against a specific disease or disorder. Experimen-tally, there are several ways to track the immune responseto indicate that a protective response is being formed. Protec-tion correlates to levels of specific Abs in blood; thus, measur-ing specific Ab response by serum ELISA is the first andfastest method of analysis. Immunization with a vaccine leadpeptide or AF will elicit populations of Abs against all compo-nents of the vaccine including the carrier protein. The peptideitself will elicit Abs that bind only to the peptide, as well asAbs that cross-react with peptide and cognate Agn. The latterare preferred, since they represent the protective response.However, Abs produced against the carrier protein are likelyto dominate the response and may also cross-react withcognate Agn (91). Serum ELISAs can reveal the proportionof the Ab response that is being diverted from the desiredtarget. Calculation of the anti-peptide to anti-carrier ratiocan indicate skewing of the Ab response toward an immuno-dominant or more prevalent component of a vaccine. Thus,optimizing a vaccine to reduce response against unwantedcomponents of an immunogen may produce a ‘‘cleaner’’ pro-tective response.

After ELISA, Abs produced by a vaccine should beanalyzed for in vitro neutralization activity or other relevantbiological activity. Since Abs may prevent or reduce infectionby different mechanisms, such as blocking of viral receptorsor complement-mediated lysis, different assays are used toconfirm such activity. The ability of an Ab to block cellularinfection by virus can be determined by plaque assay. For


example, Burton et al. (122) preincubated titrated Ab withHIV-1 virus that was then added to uninfected MT-2 cells.Cells were topped with agarose and the plates centrifugedto form cell monolayers. After several days of incubation,the plates were labeled with propidium iodide and fluorescentplaques were counted. Neutralizing titers were defined as theconcentration of Ab required to give 50–90% reduction in pla-que numbers compared to controls with no Ab.

A more recent example of this type of assay was reportedby Richman et al. (123), who used two different expressionvectors to generate HIV-1 virus-like particles (VLPs). Thefirst vector comprised the HIV-1 genome with the envelopegene replaced with a luciferase gene. The second vector borean expression cassette for intact env (which encodes gp160)from primary HIV-1 isolates. Cells that are coinfected withboth vectors make non-infectious VLPs bearing gp160 fromthe primary isolate and an HIV-1 genome that expressesluciferase instead of Env. These VLPs were then used in neu-tralization assays with sera from HIV-1 infected patients[either from the same patient that donated the envelope gene(autologous sera) or from a different patient (heterologoussera)]. Titrated sera and VLPs were coincubated with a cellline that expresses the HIV-1 receptor CD4 and the HIV-1co-receptors CCR5 and CXCR4. Neutralization activity wasmeasured as a decrease in luciferase activity in the cells afterinfection, as compared to a control.

In vitro assays are also used to test the ability of an Ab orsera to induce complement-mediated lysis of bacteria. Prinzet al. (124) immunized mice with proteosomes bearing apeptide that binds to MAb that protects from infection with aN. meningitidis serogroup C (for a similar study, see Ref. 81).A discrete cell number of N. meningitidis spirochets wasmixed with baby rabbit serum as a source of complement.Dilutions of sera from the immunized mice were mixed withthe cell=complement mixture and incubated for 1 hr. Aliquotsof the reactions were plated at time 0 and after 1 hr, andcolonies were counted after overnight incubation. The serumbactericidal titer was considered as the lowest reciprocaldilution that was required for killing 50% of the bacteria.


In vitro assays are the first way to test if a vaccine elicitsAbs with biological activity. However, protection assays arenecessary for testing the ability of a vaccine to prevent infectionin vivo. These assays follow immunization with a vaccine-leadpeptide orAFand culminate in direct exposure of the immunizedanimal to the pathogen. The results of the challengemay be ana-lyzed in several ways. The simplest is to calculate percent survi-val (56,91,102). Other methods include checking for viralclearance (101,125) or slowed disease progression (87). Depend-ing on the pathogen, challenge studies may be conducted withlive bacteria (91,124) or virus (56,87,101,102).

For example, May et al. (91) used a challenge assay todemonstrate that Abs, elicited by gluteraldehyde-treatedKLH (gKLH), did not protect mice from infection withC. neoformans, even though the Abs bound to the coat poly-saccharide, GXM. Mice, twice immunized with gKLH, weregiven an IV challenge comprising a lethal dose of C. neofor-mans strain 24067. The endpoint was number of survivinganimals; none of the mice survived. Unfortunately positivecontrols were not included to confirm the efficacy of the chal-lenge model. Another example of a bacterial challenge modelwas that used by Prinz et al. (124) (see above and below) totest the efficacy of their N. meningitidis serogroup C vaccine.Mice that were immunized three times with either pepti-de=proteosome complex or meningococcal serogroup C poly-saccharide (MCPS) were prepared for challenge by IPinjection of iron dextran, which enhances susceptibility tobacteria. After seven days, mice were given an IP challengewith 10 times the lethal dose of N. meningitidis serogroupC. The endpoint was measured as percent survival over time.After 96 hr, mice immunized with peptide=proteosome com-plex and MCPS showed 80% and 100% survival, respectively,whereas the proteosome immunized group had, zero survivalafter 36 hr. The endpoint in challenge studies is not alwaysmeasured by survival. For example, challenge studies thattest vaccine-induced immune response against RSV monitorviral clearance rather than survival (101,125). In both stu-dies, mice immunized with vaccine-lead peptides were givenIN challenge with RSV. Mice were sacrificed four or five days


postchallenge. Lung tissue was harvested, and viral clearancedetermined with plaque assays.

Passive transfer studies are also used to analyze theeffect of Abs in systems other than the ones in which theywere elicited. This has been a common approach to test theneutralizing capability of human MAbs against HIV-1 orSHIV. This has been done in monkeys (126,128,132) andhu-PBL-SCID mice (127). These types of studies may beadapted to test sera from vaccine-lead immunized animalsin other models.

The section above describes assays at several stages ofanalysis. A ‘‘vaccine lead’’ peptide or AF should satisfy all ofthe parameters discussed in this section to be considered forfurther development as a vaccine component. At the Ab level,there should exist a high level of Abs that cross-react withboth peptide or AF and cognate Agn. These Abs shoulddemonstrate biological activity in vitro (e.g., in the form ofneutralization or complement mediated lysis). However, sinceother mechanisms may contribute to the protective effects ofan Ab in vivo, challenge studies should also be performed.These results, taken together, comprise a ‘‘gold-standard’’for assessing vaccine lead peptides.

VII. SUMMARY

This chapter has reviewed the technology of phage displayand its application to diagnostics, epitope mapping, and vac-cine design. The technology and its applications have beenreviewed in the context of the three major types of libraries;whole-Agn libraries, AFLs, and RPLs. It has been the purposeof this review to outline, as well as critique, the methods andanalysis used to develop diagnostic or vaccines leads fromphage-displayed libraries. The specific focus has been ondeveloping leads for Ab-targeted vaccines using ligands fromAFLs and RPLs. As a result, there has not been an extensivediscussion on the use of whole-Agn libraries, despite theimportance of this technology. In the paragraphs below, thethree library types discussed in this chapter will be summar-ized in the context of their limitations and alternatives.


Whole-Agn libraries are mentioned, but the focus will be onAFLs and RPLs.

Whole-Agn libraries, made from the cDNA of a wholeorganism or cell line, are used to identify diagnostic orvaccine-lead proteins. The primary applications have beenfor identifying allergens or therapeutic targets (37,38,129),for isolating and identifying tumor-specific Agns (18,19,130),and for identifying Agns for autoimmune diseases such aslupus (39). The whole-Agn approach has several advantages.For example, the candidate protein is physically associatedwith its gene, and furthermore, it is possible to screenlibraries comprising many different protein Agns using highthroughput methods (37).

There are limitations to this method. For instance,cDNAs contain stop codons that prevent the expression ofopen reading frames; this can be overcome with suitablevectors (14). Moreover, bacterial expression affects transla-tion of eukaryotic proteins, specifically, protein folding andglycosylation. Yeast (131) or mammalian display systemsshould be considered as alternatives for expression of eukar-yotic cDNA.

AFLs, comprising cDNA fragments as opposed to wholecDNAs, are made from genes for single proteins (59–61) orfrom whole genomes (3). They have been used for mappinglinear epitopes (50,59,60), for identifying vaccine-lead proteinfragments (3,55) and for mapping immunodominant regionson a protein using PCAbs (50,61). AFLs are a better sourceof ligands for PCAbs or MAbs than RPLs since fragmentsare derived from cognate protein. The exception is for MAbsthat bind to discontinuous epitopes on regions of the cognateprotein that are distantly spaced or non-protein Agns such asDNA or CHO. One study has shown that AFs are more fre-quently immunogenic than peptides derived from RPLs (3).However, AFs isolated by MAbs do not always elicit Abs thatcross-react with cognate protein (55). The primary applica-tions for AFLs are epitope mapping and the identification ofligands for epitope or Ab-targeted vaccines.

Like whole-Agn libraries, AFLs are also affected byE. coli codon usage and post-translational modifications when


eukaryotic proteins are expressed. Furthermore, eachapplication requires the construction of a new library, andthis is a time-consuming task. One of the primary uses ofAFLs is epitope mapping. AFLs are well suited to mappinglinear epitopes; however, overlapping fragments need to beisolated to identify the discrete epitope rather than just thegeneral region of an epitope. There are greater problems asso-ciated with using AFLs for mapping epitopes of MAbs thatbind to discontinuous epitopes. AFLs do not often producefragments that contain discontinuous epitope regions on thesame discrete fragment. This is in spite of attempts by Huanget al. (61) to create an AFL with distant AFs ligated together.Homolog scanning, such as that done by Jin et al. (49), is analternative method for mapping discontinuous epitopes.Other options include using RPLs to define CBRs that aremapped to the cognate Agn (57,69).

RPLs are themost versatile of the library types discussed inthis chapter since they are sources of ligands forMAbs that bindto discontinuous epitopes or non-proteinAgns such asCHOs andDNA.They canbemade indifferent lengths,with orwithout con-straints, and one library may be reused for many applications.RPL technology is applicable to diagnostics, epitope mapping,and vaccine design. For example, diagnostic panels of peptides,identified with serum Abs from infected patients, characterizethe ‘‘footprint’’ of a specific disease andmay be used for diagnosis(40,41). This same approach is used for idiopathic diseases orautoimmune disorders (44,47) since RPL technology does notdepend on identification of the cognate Agn.

As mentioned above, RPLs are sources of ligands forMAbs that bind to discontinuous epitopes. Moreover, peptidesisolated from RPLs by such MAbs may be used to identify theputative epitope on the cognate protein. This approach hasbeen explored with computer-modeled epitopes for MAbs thatbind RPs (57,69). These studies show promise in the field ofdiscontinuous epitope mapping but require Ab-binding stu-dies with mutations to the putative CBRs.

Peptides derived from RPLs are commonly used invaccine development; however, they are not always optimalimmunogens. These peptides are often non-immunogenic


or do not elicit Abs that cross-react with cognate Agn(3,32,56,89). For example, Matthews et al. (3) showed thatpeptides derived from RPLs are less immunogenic than thosederived from AFLs. In addition, numerous peptides need to betested to find one that elicits Abs that cross-react with cognateAgn (56,89). However, the use of RPL-derived peptides maybe unavoidable, especially if a MAb binds a discontinuous epi-tope or a CHO Agn.

VIII. CONCLUSION

The primary aim of this chapter has been to explore the appli-cation of phage libraries to Ab-targeted vaccine design. SpecificAbs or a group of Abs against a specific epitopemay be targetedby immunization with peptides or AFs derived from RPLs andAFLs. The targeted Abs are elicited by straight immunizationwith peptide or AF, or by a prime-boost immunization, wherethe cognate Agn is used as the priming immunogen and thepeptide or AF is used as the boosting immunogen. Analysis ofthe immune response follows immunization to determinewhether the putative vaccine lead is successful or if alternativeapproaches should be explored. In the paragraphs below, thediscussion is focused on the criteria that should be evaluatedbefore declaring a vaccine-lead a success.

We propose that several criteria must be analyzedto measure the quality of Abs elicited by an Ab-targetingvaccine-lead peptide or AF. First, immunization with thevaccine-lead peptide should elicit high titers of Abs thatcross-react with the cognate Agn; these represent the propor-tionofAbs that are ‘‘protective.’’ Second, theseAbs should sharethe mechanism of biological activity with the parental Ab. Forexample, they should block against viral infection or activateADCC; this is demonstrated by in vitro neutralization assaysand in vivo challenge studies. Third, the Abs should be geneti-cally identical or very similar to the parental Ab (i.e., theyshouldhave similar geneusageand levels of somaticmutation).

The majority of the studies discussed in this chapterfocus on different aspects of Ab-targeted vaccine design. Somestudies focus on in-depth biochemical analysis of vaccine-lead


peptides whereas others focus on the immunological effect ofpeptide immunization. For example, Beenhouwer et al. (32)used a very well characterized vaccine-lead peptide and asophisticated immunization strategy that successfully elicitedhigh Ab titers against the cognate Agn. However, it was notdetermined if these Abs protected from infection, either invivo or in vitro. In contrast, the peptides isolated by Scalaet al. (51) were not optimized and were poorly characterized.Nonetheless, immunization with these peptides resulted inAbs that neutralized HIV-1 in vitro and protected macaquesfrom disease progression after infection with SHIV-89.6PD(87). A more stringent biochemical analysis of these peptideswould contribute to the knowledge of how they elicited partialprotection, and optimization of these peptides may lead to bet-ter immunogens.

In conclusion, a vaccine-lead should be considered suc-cessful when it meets both biochemical and immunologicalexpectations. In-depth biochemical analysis of a peptide leadis an excellent precursor to immunization studies but a poorpredictor of how that peptide will focus the immune responseagainst a pathogen. Conversely, a peptide that elicits a pro-tective immune response must be well characterized to ensurethat it is optimized. To date, there are no commercial peptidevaccines. Building an effective peptide vaccine will requirethe expertise of both biochemical and immunological disci-plines and a great deal of intense study.

IX. ABBREVIATIONS

Antibody (Ab), Ab-dependent cellular cytotoxicity (ADCC),antigen (Agn), Agn-fragment library (AFL), random peptidelibrary (RPL), carbohydrate (CHO), monoclonal antibody(MAb), polyclonal antibody (PCAb), critical-binding residue(CBR), open reading frame (ORF), human immunodeficiencyvirus type-1 (HIV-1), idiotype (Id), equine herpes virus (EHV),Mycoplasma capricolum subsp. capripneumoniae (MCCP),insulin-dependent diabetes mellitus (IDDM), radioimmuno-assay (RIA), Crohn’s disease (CD), human growth hormone


(huGH), hen egg lysozyme (HEL), thyroperoxidase (TPO),encephalopathogenic strain of murine hepatitis virus (E-MHV), long-term infected (LTI),Cryptococcus neoformans glu-curonoxylomannan (GXM), type-III capsular polysaccharide(type-III CPS) of group B Streptococcus (GBS), ovalbumin(OVA), bovine serum albumin (BSA), keyhole limpethemocyanin (KLH), meningococcal group A polysaccharide(MGAPS), B-cell epitope (BCE), TH-cell epitope (TCE), tetanustoxoid (TT), hepatitisB surfaceAgn (HbsAgn), hepatitisBviruscore peptide (HBVcore), respiratory syncytial virus(RSV), herpes simplex virus (HSV), Epstein–Barr virus(EBV), endoplasmic reticulum(ER), cytotoxicT-cell (CTL), sub-cutaneous (SC), toxic shock syndrome toxin-1 (TSST-1), outermembrane porin protein (OprF), Agn presenting cell (APC),monophosphoryl lipid A (MPL), granulocyte-macrophage col-ony stimulating factor (GM-CSF), intramuscular (IM), intra-peritoneal (IP), oral (PO), intra-nasal (IN), dendritic cells(DCs), virus-like particle (VLP), gluteraldehyde-treated KLH(gKLH), meningococcal serogroup C polysaccharide (MCPS),and multiple antigenic peptide (MAP).

REFERENCES

1. Janeway CA, Travers P, Walport M, Shlomchik M. Immuno-biology: the Immune System in Health and Disease. 5th ed.New York: Garland Publishing, 2001.

2. Craig L, Sanschagrin PC, Rozek A, Lackie S, Kuhn LA, ScottJK. The role of structure in antibody cross-reactivity betweenpeptides and folded proteins. J Mol Biol 1998; 281:183–201.

3. Matthews LJ, Davis R, Smith GP. Immunogenically fit subu-nit vaccine components via epitope discovery from naturalpeptide libraries. J Immunol 2002; 169:837–846.

4. Wells JA. Binding in the growth hormone receptor complex.Proc Natl Acad Sci USA 1996; 93:1–6.

5. Barlow DJ, Edwards MS, Thornton JM. Continuous anddiscontinuous protein antigenic determinants. Nature 1986;322:747–748.


6. Geysen HM, Tainer JA, Rodda SJ, Mason TJ, Alexander H,Getzoff ED, Lerner RA. Chemistry of antibody binding to aprotein. Science 1987; 235:1184–1190.

7. Davies DR, Cohen GH. Interactions of protein antigens withantibodies. Proc Natl Acad Sci USA 1996; 93:7–12.

8. Harris SL, Craig L, Mehroke JS, Rashed M, Zwick MB, KenarK, Toone EJ, Greenspan N, Auzanneau FI, Marino-AlbernasJR, Pinto BM, Scott JK. Exploring the basis of peptide-carbohydrate crossreactivity: evidence for discrimination bypeptides between closely related anti-carbohydrate antibo-dies. Proc Natl Acad Sci USA 1997; 94:2454–2459.

9. Vyas NK, Vyas MN, Chervenak MC, Johnson MA, Pinto BM,Bundle DR, Quiocho FA. Molecular recognition of oligosac-charide epitopes by a monoclonal Fab specific for Shigellaflexneri Y lipopolysaccharide: x-ray structures and thermody-namics. Biochemistry 2002; 41:13575–13586.

10. Vyas NK, Vyas MN, Chervenak MC, Bundle DR, Pinto BM,Quiocho FA. Structural basis of peptide-carbohydrate mimi-cry in an antibody-combining site. Proc Natl Acad Sci USA2003; 100:15023–15028.


12. Zhong G. Conformational mimicry through random con-straints plus affinity selection. Methods Mol Biol 1998; 87:165–173.

13. Barbas CF III, Burton DR, Scott JK, Silverman GJ. PhageDisplay: a Laboratory Manual. Cold Spring Harbor: ColdSpring Harbor Laboratory Press, 2001.

14. CrameriR,SuterM.Display of biologically activeproteins on thesurface of filamentous phages: a cDNA cloning system for selec-tion of functional geneproducts linked to the genetic informationresponsible for their production. Gene 1993; 137:69–75.

15. Jespers LS, Messens JH, De Keyser A, Eeckhout D, Van denBrande I, Gansemans YG, Lauwereys MJ, Vlasuk GP,Stanssens PE. Surface expression and ligand-based selectionof cDNAs fused to filamentous phage gene VI. Biotechnology(NY) 1995; 13:378–382.


16. Hufton SE, Moerkerk PT, Meulemans EV, de Bruine A,Arends JW, Hoogenboom HR. Phage display of cDNA reper-toires: the pVI display system and its applications for theselection of immunogenic ligands. J Immunol Methods1999; 231:39–51.

17. Jacobsson K, Frykberg L. Gene VIII-based, phage-displayvectors for selection against complex mixtures of ligands.Biotechniques 1998; 24:294–301.

18. Somers VA, Brandwijk RJ, Joosten B, Moerkerk PT, ArendsJW, Menheere P, Pieterse WO, Claessen A, Scheper RJ,Hoogenboom HR, Hufton SE. A panel of candidate tumorantigens in colorectal cancer revealed by the serological selec-tion of a phage displayed cDNA expression library. J Immu-nol 2002; 169:2772–2780.

19. Sioud M, Hansen MH. Profiling the immune responsein patients with breast cancer by phage-displayed cDNAlibraries. Eur J Immunol 2001; 31:716–725.

20. AJ van Zonneveld, van den Berg BM, van Meijer M,Pannekoek H. Identification of functional interaction siteson proteins using bacteriophage-displayed random epitopelibraries. Gene 1995; 167:49–52.

21. Zacchi P, Sblattero D, Florian F, Marzari R, Bradbury AR.Selecting open reading frames from DNA. Genome Res 2003;13:980–990.





26. Virnekaes B, Ge L, Plueckthun A, Schneider KC, WellnhoferG, Moroney SE. Trinucleotide phosphoramidites: ideal


reagents for the synthesis of mixed oligonucleotides forrandom mutagenesis. Nucleic Acids Res 1994; 22:5600–5607.

27. Menendez A, Bonnycastle LL, Pan OCC, Scott JK. Screeningpeptide libraries. In: Barbas CF III, Burton DR, Scott JK,Silverman GJ, eds. Phage Display: a Laboratory Manual.Cold Spring Harbor: Cold Spring Harbor Laboratory Press,2001:17.11–17.32.

28. Kishchenko G, Batliwala H, Makowski L. Structure of aforeign peptide displayed on the surface of bacteriophageM13. J Mol Biol 1994; 241:208–213.

29. Zwick MB, Bonnycastle LL, Menendez A, Irving MB, BarbasCF III, Parren PW, Burton DR, Scott JK. Identification andcharacterization of a peptide that specifically binds thehuman, broadly neutralizing anti-human immunodeficiencyvirus type 1 antibody b12. J Virol 2001; 75:6692–6699.

30. Wrighton NC, Farrell FX, Chang R, Kashyap AK, BarboneFP, Mulcahy LS, Johnson DL, Barrett RW, Jolliffe LK, DowerWJ. Small peptides as potent mimetics of the protein hor-mone erythropoietin. Science 1996; 273:458–464.

31. Zhu ZY, Minenkova O, Bellintani F, De Tomassi A, UrbanelliL, Felici F, Monaci P. ‘In vitro evolution’ of ligands for HCV-specific serum antibodies. Biol Chem 2000; 381:245–254.

32. Beenhouwer DO, May RJ, Valadon P, Scharff MD. High affi-nity mimotope of the polysaccharide capsule of Cryptococcusneoformans identified from an evolutionary phage peptidelibrary. J Immunol 2002; 169:6992–6999.

33. Glaser SM, Yelton DE, Huse WD. Antibody engineering bycodon-based mutagenesis in a filamentous phage vectorsystem. J Immunol 1992; 149:3903–3913.

34. Marston EL, James AV, Parker JT, Hart JC, Brown TM,Messmer TO, Jue DL, Black CM, Carlone GM, Ades EW,Sampson J. Newly characterized species-specific immuno-genic Chlamydophila pneumoniae peptide reactive with mur-ine monoclonal and human serum antibodies. Clin Diagn LabImmunol 2002; 9:446–452.

35. Crameri R, Jaussi R, Menz G, Blaser K. Display of exp-ressionproducts of cDNAlibraries onphage surfaces.A versatile


screening system for selective isolation of genes by specific gene-product=ligand interaction. Eur J Biochem 1994; 226:53–58.

36. Crameri R, Faith A, Hemmann S, Jaussi R, Ismail C, Menz G,Blaser K. Humoral and cell-mediated autoimmunity in allergyto Aspergillus fumigatus. J Exp Med 1996; 184:265–270.

37. Weichel M, P Schmid-Grendelmeier, Rhyner C, Achatz G,Blaser K, Crameri R. Immunoglobulin E-binding and skintest reactivity to hydrophobin HCh-1 from Cladosporiumherbarum, the first allergenic cell wall component of fungi.Clin Exp Allergy 2003; 33:72–77.

38. Rhyner C, Kodzius R, Crameri R. Direct selection of cDNAsfrom filamentous phage surface display libraries: potentialand limitations. Curr Pharm Biotechnol 2002; 3:13–21.

39. Kemp EH, Herd LM,Waterman EA,Wilson AG,Weetman AP,Watson PP. Immunoscreening of phage-displayed cDNA-encoded polypeptides identifies B-cell targets in autoimmunedisease. Biochem Biophys Res Commun 2002; 298:169–177.

40. Folgori A, Tafi R, Meola A, Felici F, Galfre G, Cortese R,Monaci P, Nicosia A. A general strategy to identify mimo-topes of pathological antigens using only random peptidelibraries and human sera. EMBO J 1994; 13:2236–2243.

41. Kouzmitcheva GA, Petrenko VA, Smith GP. Identifying diag-nostic peptides for lyme disease through epitope discovery.Clin Diagn Lab Immunol 2001; 8:150–160.

42. Birch-Machin I, Ryder S, Taylor L, Iniguez P, Marault M,Ceglie L, Zientara S, Cruciere C, Cancellotti F, KoptopoulosG, Mumford J, Binns M, Davis-Poynter N, Hannant D. Utili-sation of bacteriophage display libraries to identify peptidesequences recognised by equine herpesvirus type 1 specificequine sera. J Virol Methods 2000; 88:89–104.

43. Benguric DR, Dungu B, Thiaucourt F, du Plessis DH. Phagedisplayed peptides and anti-idiotype antibodies recognised bya monoclonal antibody directed against a diagnostic antigenof Mycoplasma capricolum subsp. capripneumoniae. VetMicrobiol 2001; 81:165–179.

44. Fierabracci A, Biro PA, Yiangou Y, Mennuni C, Luzzago A,Ludvigsson J, Cortese R, Bottazzo GF. Osteopontin is an


autoantigen of the somatostatin cells in human islets: identi-fication by screening random peptide libraries with sera ofpatients with insulin-dependent diabetes mellitus. Vaccine1999; 18:342–354.

45. Yamagishi S, Fujimori H, Yonekura H, Tanaka N, YamamotoH. Advanced glycation endproducts accelerate calcificationin microvascular pericytes. Biochem Biophys Res Commun1999; 258:353–357.

46. Chen NX, Moe SM. Arterial calcification in diabetes. CurrDiab Rep 2003; 3:28–32.

47. Saito H, Fukuda Y, Katsuragi K, Tanaka M, Satomi M,Shimoyama T, Saito T, Tachikawa T. Isolation of peptidesuseful for differential diagnosis of Crohn’s disease and ulcera-tive colitis. Gut 2003; 52:535–540.

48. Visvanathan S, Scott JK, Hwang KK, Banares M, GrossmanJM, Merrill JT, FitzGerald J, Chukwuocha RU, Tsao BP,Hahn BH, Chen PP. Identification and characterization of apeptide mimetic that may detect a species of disease-associated anticardiolipin antibodies in patients with theantiphospholipid syndrome. Arthritis Rheum 2003; 48:37–745.

49. Jin L, Fendly BM, Wells JA. High resolution functionalanalysis of antibody–antigen interactions. J Mol Biol 226:851–865.

50. Bentley L, Fehrsen J, Jordaan F, Huismans H, du PlessisDH. Identification of antigenic regions on VP2 of Africanhorsesickness virus serotype 3 by using phage-displayedepitope libraries. J Gen Virol 2000; 81:993–1000.

51. Scala G, Chen X, Liu W, Telles JN, Cohen OJ, Vaccarezza M,Igarashi T, Fauci AS. Selection of HIV-specific immunogenicepitopes by screening random peptide libraries with HIV-1-positive sera. J Immunol 1999; 162:6155–6161.

52. Enshell-Seijffers D, Smelyanski L, Vardinon N, Yust I,Gershoni JM. Dissection of the humoral immune responsetoward an immunodominant epitope of HIV: a model for theanalysis of antibody diversity in HIVþ individuals. FASEBJ 2001; 15:2112–2120.


53. Bonnycastle LL, Mehroke JS, Rashed M, Gong X, Scott JK.Probing the basis of antibody reactivity with a panel of con-strained peptide libraries displayed by filamentous phage. JMol Biol 1996; 258:747–762.

54. Zhu Z, Ming Y, Sun B. Identification of epitopes of tricho-santhin by phage peptide library. Biochem Biophys ResCommun 2001; 282:921–927.

55. Yip YL, Smith G, Koch J, Dubel S, Ward RL. Identification ofepitope regions recognized by tumor inhibitory and stimula-tory anti-ErbB-2 monoclonal antibodies: implications forvaccine design. J Immunol 2001; 166:5271–5278.

56. Yu MW, Scott JK, Fournier A, Talbot PJ. Characterization ofmurine coronavirus neutralization epitopes with phage-displayed peptides. Virology 2000; 271:182–196.

57. Bresson D, Cerutti M, Devauchelle G, Pugniere M, Roquet F,Bes C, Bossard C, Chardes T, Peraldi-Roux S. Localization ofthe discontinuous immunodominant region recognized byhuman anti-thyroperoxidase autoantibodies in autoimmunethyroid diseases. J Biol Chem 2003; 278:9560–9569.

58. Parhami-Seren B, Keel T, Reed GL. Sequences of antigenicepitopes of streptokinase identified via random peptidelibraries displayed on phage. J Mol Biol 1997; 271:333–341.

59. Fack F, Hugle-Dorr B, Song D, Queitsch I, Petersen G, BautzEK. Epitope mapping by phage display: random versus gene-fragment libraries. J Immunol Methods 1997; 206:43–52.

60. Holzem A, Nahring JM, Fischer R. Rapid identification of atobacco mosaic virus epitope by using a coat protein gene-fragment–pVIII fusion library. J Gen Virol 2001; 82:9–15.

61. Huang JA, Wang L, Firth S, Phelps A, Reeves P, Holmes I.Rotavirus VP7 epitope mapping using fragments of VP7 dis-played on phages. Vaccine 2000; 18:2257–2265.

62. Stemmer WP. Rapid evolution of a protein in vitro by DNAshuffling. Nature 1994; 370:389–391.

63. Ollman Saphire E, Parren PW, Pantophlet R, Zwick MB,Morris GM, Rudd PM, Dwek RA, Stanfield RL, Burton DR,Wilson IA. Crystal structure of a neutralizing human IGG


against HIV-1: a template for vaccine design. Science 2001;293:1155–1159.

64. Stern B, Denisova G, Buyaner D, Raviv D, Gershoni JM.Helical epitopes determined by low-stringency antibodyscreening of a combinatorial peptide library. FASEB J 1997;11:147–153.

65. Kwong PD, Wyatt R, Robinson J, Sweet RW, Sodroski J,Hendrickson WA. Structure of an HIV gp120 envelope glyco-protein in complex with the CD4 receptor and a neutralizinghuman antibody. Nature 1998; 393:648–659.

66. Ferrieres G, Villard S, Pugniere M, Mani JC, Navarro-TeulonI, Rharbaoui F, Laune D, Loret E, Pau B, Granier C. Affinityfor the cognate monoclonal antibody of synthetic peptidesderived from selection by phage display. Role of sequencesflanking the binding motif. Eur J Biochem 2000; 267:1819–1829.

67. Zwick MB, Bonnycastle LL, Noren KA, Venturini S, Leong E,Barbas CF III, Noren CJ, Scott JK. The maltose-binding pro-tein as a scaffold for monovalent display of peptides derivedfrom phage libraries. Anal Biochem 1998; 264:87–97.

68. Menendez A, Chow KC, Pan O, Scott JK. Human immunode-ficiency virus type 1-neutralizing monoclonal antibody 2F5 ismultispecific for sequences flanking the DKW core epitope. JMol Biol 2004; 338:311–327.

69. Enshell-Seijffers D, Denisov D, Groisman B, Smelyanski L,Meyuhas R, Gross G, Denisova G, Gershoni JM. The mappingand reconstitution of a conformational discontinuous B-cellepitope of HIV-1. J Mol Biol 2003; 334:87–101.

70. Pantophlet R, Wilson IA, Burton DR. Hyperglycosylatedmutants of human immunodeficiency virus (HIV) type 1mono-meric gp120 as novel antigens for HIV vaccine design. J Virol2003; 77:5889–5901.

71. Pantophlet R, Ollmann Saphire E, Poignard P, Parren PW,Wilson IA, Burton DR. Fine mapping of the interaction ofneutralizing and nonneutralizing monoclonal antibodies withthe CD4 binding site of human immunodeficiency virus type 1gp120. J Virol 2003; 77:642–658.


72. Plotkin SA. Vaccines, vaccination, and vaccinology. J InfectDis 2003; 187:1349–1359.

73. Moylett EH, Hanson IC. 29. Immunization. J Allergy ClinImmunol 2003; 111:S754–S765.

74. Jerne NK. Towards a network theory of the immune system.Ann Immunol (Paris) 1974; 125C:373–389.

75. Goldbaum FA, Velikovsky CA, Dall’Acqua W, Fossati CA,Fields BA, Braden BC, Poljak RJ, Mariuzza RA. Characteriza-tion of anti–anti-idiotypic antibodies that bind antigen and ananti-idiotype. Proc Natl Acad Sci USA 1997; 94:8697–8701.

76. Hutchins WA, Adkins AR, Kieber-Emmons T, Westerink MA.Molecular characterization of a monoclonal antibody pro-duced in response to a group C meningococcal polysaccharidepeptide mimic. Mol Immunol 1996; 33:503–510.

77. Westerink MA, Campagnari AA, Wirth MA, Apicella MA.Development and characterization of an anti-idiotype anti-body to the capsular polysaccharide of Neisseria meningitidisserogroup C. Infect Immun 1988; 56:1120–1127.

78. Caton M, Diamond B. Using peptide mimetopes to elucidateanti-polysaccharide and anti-nucleic acid humoral responses.Cell Mol Biol (Noisy-le-grand) 2003; 49:255–262.

79. Benvenuti F, Cesco-Gaspere M, Burrone OR. Anti-idiotypicDNA vaccines for B-cell lymphoma therapy. Front Biosci2002; 7:d228–d234.

80. Bendandi M. Anti-idiotype vaccines for human follicularlymphoma. Leukemia 2000; 14:1333–1339.

81. Westerink MA, Smithson SL, Hutchins WA, Widera G. Devel-opment and characterization of anti-idiotype based peptideand DNA vaccines which mimic the capsular polysaccharideof Neisseria meningitidis serogroup C. Int Rev Immunol2001; 20:251–261.

82. Kieber-Emmons T, Getzoff E, Kohler H. Perspectives onantigenicity and idiotypy. Int Rev Immunol 1987; 2:339–356.

83. Fields BA, Goldbaum FA, Ysern X, Poljak RJ, Mariuzza RA.Molecular basis of antigen mimicry by an anti-idiotope.Nature 1995; 374:739–742.


84. Putterman C, Deocharan B, Diamond B. Molecular analysisof the autoantibody response in peptide-induced autoimmu-nity. J Immunol 2000; 164:2542–2549.

85. Ganglberger E, Grunberger K, Sponer B, Radauer C,Breiteneder H, Boltz-Nitulescu G, Scheiner O, Jensen-Jarolim E. Allergen mimotopes for 3-dimensional epitopesearch and induction of antibodies inhibiting human IgE.FASEB J 2000; 14:2177–2184.

86. Frenkel D, Katz O, Solomon B. Immunization againstAlzheimer’s beta-amyloid plaques via EFRH phage adminis-tration. Proc Natl Acad Sci USA 2000; 97:11455–11459.

87. Chen X, Scala G, Quinto I, Liu W, Chun TW, Justement JS,Cohen OJ, vanCott TC, Iwanicki M, Lewis MG, GreenhouseJ, Barry T, Venzon D, Fauci AS. Protection of rhesus maca-ques against disease progression from pathogenic SHIV-89.6PD by vaccination with phage-displayed HIV-1 epitopes.Nat Med 2001; 7:1225–1231.

88. Lesinski GB, Westerink MA. Novel vaccine strategies toT-independent antigens. J Microbiol Methods 2001; 47:135–149.

89. Phalipon A, Folgori A, Arondel J, Sgaramella G, Fortugno P,Cortese R, Sansonetti PJ, Felici F. Induction of anti-carbohydrate antibodies by phage library-selected peptidemimics. Eur J Immunol 1997; 27:2620–2625.

90. Pincus SH, Smith MJ, Jennings HJ, Burritt JB, Glee PM.Peptides that mimic the group B streptococcal type III cap-sular polysaccharide antigen. J Immunol 1998; 160:293–298.

91. May RJ, Beenhouwer DO, Scharff MD. Antibodies to keyholelimpet hemocyanin cross-react with an epitope on thepolysaccharide capsule of Cryptococcus neoformans and othercarbohydrates: implications for vaccine development. JImmunol 2003; 171:4905–4912.

92. JohnsonMA,JasejaM,ZouW,JenningsHJ,CopieV,PintoBM,Pincus SH. NMR studies of carbohydrates and carbohydrate-mimetic peptides recognized by an anti-group B Streptococcusantibody. J Biol Chem 2003; 278:24740–24752.


93. Grothaus MC, Srivastava N, Smithson SL, Kieber-EmmonsT, Williams DB, Carlone GM, Westerink MA. Selection ofan immunogenic peptide mimic of the capsular polysacchar-ide of Neisseria meningitidis serogroup A using a peptidedisplay library. Vaccine 2000; 18:1253–1263.

94. Lohnas GL, Roberts SF, Pilon A, Tramontano A. Epitope-specific antibody and suppression of autoantibody responsesagainst a hybrid self protein. J Immunol 1998; 161:6518–6525.

95. Cox JC, Coulter AR. Adjuvants—a classification and reviewof their modes of action. Vaccine 1997; 15:248–256.

96. Edelman R. The development and use of vaccine adjuvants.Mol Biotechnol 2002; 21:129–148.

97. Tam JP. Synthetic peptide vaccine design: synthesis andproperties of a high-density multiple antigenic peptidesystem. Proc Natl Acad Sci USA 1988; 85:5409–5413.

98. Coombes AG, Lavelle EC, Jenkins PG, Davis SS. Single dose,polymeric, microparticle-based vaccines: the influence of for-mulation conditions on the magnitude and duration of theimmune response to a protein antigen. Vaccine 1996; 14:1429–1438.

99. de la Cruz VF, Lal AA, McCutchan TF. Immunogenicity andepitope mapping of foreign sequences via genetically engi-neered filamentous phage. J Biol Chem 1988; 263:4318–4322.

100. Meola A, Delmastro P, Monaci P, Luzzago A, Nicosia A, FeliciF, Cortese R, Galfre G. Derivation of vaccines from mimo-topes. Immunologic properties of human hepatitis B virussurface antigen mimotopes displayed on filamentous phage.J Immunol 1995; 154:3162–3172.

101. Bastien N, Trudel M, Simard C. Protective immuneresponses induced by the immunization of mice with a recom-binant bacteriophage displaying an epitope of the humanrespiratory syncytial virus. Virology 1997; 234:118–122.

102. Grabowska AM, Jennings R, Laing P, Darsley M, JamesonCL, Swift L, Irving WL. Immunisation with phage displayingpeptides representing single epitopes of the glycoprotein G


can give rise to partial protective immunity to HSV-2.Virology 2000; 269:47–53.

103. Kneissel S, Queitsch I, Petersen G, Behrsing O, Micheel B,Dubel S. Epitope structures recognised by antibodies againstthe major coat protein (g8p) of filamentous bacteriophage fd(Inoviridae). J Mol Biol 1999; 288:21–28.

104. Willis AE, Perham RN, Wraith D. Immunological propertiesof foreign peptides in multiple display on a filamentousbacteriophage. Gene 1993; 128:79–83.

105. De Berardinis P, Sartorius R, Fanutti C, Perham RN, DelPozzo G, Guardiola J. Phage display of peptide epitopes fromHIV-1 elicits strong cytolytic responses. Nat Biotechnol 2000;18:873–876.

106. De Berardinis P, D’Apice L, Prisco A, Ombra MN, Barba P,Del Pozzo G, Petukhov S, Malik P, Perham RN, GuardiolaJ. Recognition of HIV-derived B and T-cell epitopes displayedon filamentous phages. Vaccine 1999; 17:1434–1441.

107. Manoutcharian K, Terrazas LI, Gevorkian G, Acero G,Petrossian P, Rodriguez M, Govezensky T. Phage-displayedT-cell epitope grafted into immunoglobulin heavy-chaincomplementarity-determining regions: an effective vaccinedesign tested in murine cysticercosis. Infect Immun 1999;67:4764–4770.

108. Petrenko VA, Smith GP, Gong X, Quinn T. A library oforganic landscapes on filamentous phage. Protein Eng 1996;9:797–801.


110. Felici F, Castagnoli L, Musacchio A, Jappelli R, Cesareni G.Selection of antibody ligands from a large library of oligopep-tides expressed on a multivalent exposition vector. J Mol Biol1991; 222:301–310.

111. Zwick MB, Shen J, Scott JK. Homodimeric peptides displayedby the major coat protein of filamentous phage. J Mol Biol2000; 300:307–320.


112. Monette M, Opella SJ, Greenwood J, Willis AE, Perham RN.Structure of a malaria parasite antigenic determinant dis-played on filamentous bacteriophage determined by NMRspectroscopy: implications for the structure of continuouspeptide epitopes of proteins. Protein Sci 2001; 10:1150–1159.

113. Yip YL, Smith G, Ward RL. Comparison of phage pIII, pVIIIand GST as carrier proteins for peptide immunisation inBalb=c mice. Immunol Lett 2001; 79:197–202.

114. Rubinchik E, Chow AW. Recombinant expression and neutra-lizing activity of an MHC class II binding epitope of toxicshock syndrome toxin-1. Vaccine 2000; 18:2312–2320.

115. Kato H, Bukawa H, Hagiwara E, Xin KQ, Hamajima K,Kawamoto S, Sugiyama M, Noda E, Nishizaki M, Okuda K.Rectal and vaginal immunization with a macromolecularmulticomponent peptide vaccine candidate for HIV-1 infec-tion induces HIV-specific protective immune responses.Vaccine 2000; 18:1151–1160.

116. Bennett B, Check IJ, Olsen MR, Hunter RL. A comparison ofcommercially available adjuvants for use in research. JImmunol Methods 1992; 153:31–40.

117. Johnston BA, Eisen H, Fry D. An evaluation of severaladjuvant emulsion regimens for the production of polyclonalantisera in rabbits. Lab Anim Sci 1991; 41:15–21.

118. Kenney JS, Hughes BW, Masada MP, Allison AC. Influenceof adjuvants on the quantity, affinity, isotype and epitopespecificity of murine antibodies. J Immunol Methods 1989;121:157–166.

119. Zuercher AW,Miescher SM, Vogel M, Rudolf MP, Stadler MB,Stadler BM. Oral anti-IgE immunization with epitope-displaying phage. Eur J Immunol 30:128–135.

120. Delmastro P, Meola A, Monaci P, Cortese R, Galfre G.Immunogenicity of filamentous phage displaying peptidemimotopes after oral administration. Vaccine 1997; 15:1276–1285.

121. Holmgren J, Czerkinsky C, Eriksson K, Mharandi A. Muco-sal immunisation and adjuvants: a brief overview of recentadvances and challenges. Vaccine 2003; 21(suppl 2):S89–S95.


122. Burton DR, Pyati J, Koduri R, Sharp SJ, Thornton GB,Parren PW, Sawyer LS, Hendry RM, Dunlop N, Nara PL,et al. Efficient neutralization of primary isolates of HIV-1by a recombinant human monoclonal antibody. Science1994; 266:1024–1027.

123. Richman DD, Wrin T, Little SJ, Petropoulos CJ. Rapid evolu-tion of the neutralizing antibody response to HIV type 1 infec-tion. Proc Natl Acad Sci USA 2003; 100:4144–4149.

124. Prinz DM, Smithson SL, Westerink MA. Two different meth-ods result in the selection of peptides that induce a protectiveantibody response to Neisseria meningitidis serogroup C.J Immunol Methods 2004; 285:1–14.

125. Chargelegue D, Obeid OE, Hsu SC, Shaw MD, Denbury AN,Taylor G, Steward MW. A peptide mimic of a protective epi-tope of respiratory syncytial virus selected from a combina-torial library induces virus-neutralizing antibodies andreduces viral load in vivo. J Virol 1998; 72:2040–2046.

126. Mascola JR, Stiegler G, VanCott TC, Katinger H, CarpenterCB, Hanson CE, Beary H, Hayes D, Frankel SS, Birx DL,Lewis MG. Protection of macaques against vaginal trans-mission of a pathogenic HIV-1=SIV chimeric virus bypassive infusion of neutralizing antibodies. Nat Med 2000;6:207–210.

127. Gauduin MC, Parren PW, Weir R, Barbas CF, Burton DR,Koup RA. Passive immunization with a human mon-oclonal antibody protects hu-PBL-SCID mice against chal-lenge by primary isolates of HIV-1. Nat Med 1997; 3:1389–1393.

128. Parren PW, Marx PA, Hessell AJ, Luckay A, Harouse J,Cheng-Mayer C, Moore JP, Burton DR. Antibody protectsmacaques against vaginal challenge with a pathogenic R5simian=human immunodeficiency virus at serum levelsgiving complete neutralization in vitro. J Virol 2001; 75:8340–8347.

129. Crameri R, Blaser K. Cloning Aspergillus fumigatus aller-gens by the pJuFo filamentous phage display system. IntArch Allergy Immunol 1996; 110:41–45.


130. Rotblat B,Enshell-SeijffersD,Gershoni JM, Schuster S,AvniA.Identification of an essential component of the elicitation activesite of the EIX protein elicitor. Plant J 2002; 32:1049–1055.

131. Boder ET, Wittrup KD. Yeast surface display for screeningcombinatorial polypeptide libraries. Nat Biotechnol 1997; 15:553–557.

132. Baba TW, Liska V, Hofmann-Lehmann R, Vlasak J, Xu W,Ayehunie S, Cavacini LA, Posner MR, Katinger H, StieglerG, Bernacky BJ, Rizvi TA, Schmidt R, Hill LR, Keeling ME,Lu Y, Wright JE, Chou TC, Ruprecht RM. Human neutraliz-ing monoclonal antibodies of the IgG1 subtype protectagainst mucosal simian-human immunodeficiency virusinfection. Nat Med 2000; 6: 200–206.


6

Exploring Protein–ProteinInteractions Using Peptide Libraries

Displayed on Phage

KURT DESHAYES

Department of ProteinEngineering,Genentech Inc.,South San Francisco, California, U.S.A.

I. INTRODUCTION

It has become evident that protein–protein interactions playa central role in signal transduction, and are thus key regu-lators of cell function. It has also become clear that theidentification of therapeutically important protein–proteininteractions from among the thousands of contenders req-uires rapid and robust screening methodologies. Fortunately,the display of naıve peptide libraries on phage has provenan effective tool for the exploration of binding surfacesand the discovery of novel binding partners (1). The

255

utility of phage display is demonstrated by the repeatedsuccess of phage sorting in yielding potent binders againstproteins for which other methods fail to discover specificligands.

II. EXTRACELLULAR PROTEIN–PROTEININTERACTIONS

Small molecule screening efforts against extracellularprotein–protein interactions generally fail to identify antago-nists, presumably because extracellular protein binding sitesare presented over large areas without significant invagina-tion (2). In contrast, phage-displayed peptide libraries consis-tently yield binders that recognize features presented overlarge surface areas.

The ability of phage display to discover potent peptidereagents has enabled the detailed elucidation of biologicalrecognition in several systems. As illustrated in Fig. 1, pep-tide binders selected against extracellular targets con-sistently contain internal disulfide bonds. The constraintsimposed by disulfide bonds stabilize peptide structure andorganize the binding surfaces. The consistent selection ofstructured peptides (3–9) that bind to extracellular proteinssuggests that preorganization of the binding surface isrequired for high-affinity binding. Peptides that bind to avariety of targets frequently have either b-hairpin or turn-helix structures, but other folds are also observed (Fig. 1)(4–15).

The interactions between phage-derived peptide andreceptor often lead to observable changes in function. Fre-quently, the phage-derived peptide competes with the naturalligand for overlapping binding sites (4–14,16). This observa-tion leads one to surmise that proteins have evolved regionsthat function as definitive binding sites. The characteristicsof these binding sites are an important issue when evaluatingpotential binding partners, or evaluating the possibility ofsmall molecule therapeutics. This chapter presents examplesthat illustrate how phage-displayed peptide libraries can be

256 Deshayes

used to assess the physical characteristics of protein–proteininteractions.

II.A. Erythropoietin Receptor

An important example of efficacy of phage display is thediscovery of peptides that either agonize or antagonize theprimary pathway for red blood cell proliferation (10,11). Poly-valent phage display methods were used to identify the20-residue peptide EMP1, which binds tightly to erythropoie-tin receptor (EPOR) (Fig. 2) (10).

Not only does EMP1 contain an internal disulfide bondthat stabilizes the b-hairpin structure, it also exists as a

Figure 1 Selected phage-derived peptide antagonists andagonists for which structures have been determined bound to, orfree of, their target proteins. The name of the target protein isshown above each peptide. Secondary structural elements areshown as ribbons along with the disulfide bonds. The literaturereferences are as follows: BR3 (3), IgG-Fc (13), EPOR (10), VEGF(15,16), IGFBP-1 (4), FVIIa (5), and IGF-1 (9).

Phage Peptide Libraries and Protein Detection 257

homodimer in solution, with a 320 A2 interface between themonomers. EMP1 blocks binding of erythropoietin to EPORwith an IC50 of 200 nM. Structural analysis shows that theEMP1 dimer forms a 1720 A2 interface that simultaneouslycontacts two EPOR molecules (Fig. 3).

Binding of the EMP1 dimer activates the erythropoietin-mediated signaling pathway, as demonstrated by the prolif-eration of erythroid cells (10). The initial EMP1 results agreewith the model in which dimerization of isolated EPOR mole-cules brings together associated kinase domains residingacross the cellular membrane (17). Interactions between thekinase domains initiate a series of events that transmit a sig-nal into the nucleus. Amazingly, substitution of the tyrosineresidue in EMP1 with 3,5-dibromotyrosine converts theagonist into an antagonist (EMP33) (11). Structural analysisreveals that EMP33 is also a dimer, structurally identicalto EMP1 except for deviations at both termini (18). Thesevariations in peptide structure are sufficient to reorient thetwo EPOR molecules within the complex, preventing therequisite interaction between associated kinases (Fig. 3).

Comparison of the EMP1–EPOR and EMP33–EPORstructures shows that subtle changes in EPOR orientationdistinguish an inactive structure. Subsequent studies pro-vided evidence that two EPORs exist in an inactive preformed

Figure 2 Solution structure of the EMP1 dimer.

258 Deshayes

dimer, and signaling through EPOR occurs upon alignment oftwo preassociated EPOR molecules into an activated state(18,19). This is distinct from the mechanisms of other cyto-kines, such as human growth hormone, in which dimerizationof two independent receptors is required for signaling (17).The relative generality of these two mechanisms has not beenestablished, but the observation that the mechanism of cyto-kine signal transduction varies among different pathwayshas far-reaching mechanistic implications.

II.B. Insulinlike Growth Factor

The recent solution structure of insulin-like growth factor 1(IGF-1) is a good example of how peptide phage display canprovide reagents that can be used to provide crucial struc-tural information unavailable through structural analysis of

Figure 3 Structure of the complex between the EMP1 dimer andthe erythropoietin receptor dimer. The peptide is shown as a ribbon.


the receptor alone (9,20). IGF-1 is a 70-residue peptidehormone that regulates both mitogenic and metabolic func-tions, primarily via binding to the cell surface IGF-1 receptor(IGFR), a cell surface receptor related to the insulin receptor(IR) (21). IGF-1 activity is tightly regulated, existing in vivoprimarily bound to one of six soluble IGF-binding proteins(IGFBPs) that sequester the hormone within high-affinitycomplexes, thereby controlling IGF-1 biological availabilityand plasma half-life (22).

Nuclear magnetic resonance studies of IGF-1 in solutiongave poor results due to self-association and unfavorableinternal dynamics (23,24). The situation changes markedlyupon the addition of the phage-derived IGF-1 ligand F1-1.F1-1 is a 17-residue peptide that contains a single disulfidebond and inhibits IGF-1 binding to four receptors with IC50

values in the low micromolar range (9). Interestingly, IGF-1becomes highly structured upon binding F1-1 (Fig. 4), adopt-ing a conformation similar to that of insulin (20). This resultclearly demonstrates that IGF-1 only adopts a stable confor-mation when bound to an ancillary molecule (20,25,26).

Comparison of the solution structure of the IGF-1 F1-1complex with crystal structures of IGF-1 bound to either afragment of IGFBP-5 or a detergent molecule reveals thatthe same hydrophobic patch on IGF-1 makes contactwith each ligand (20). The differences in IGF-1 structure incomplexes are due to changes in helix 3 conformation thatchange the orientation of important contact residues. Forexample, when IGF-1 binds a detergent molecule, helix 3begins as a 310 helix but ends as a regular a-helix (27). In con-trast, when F1-1 is bound to IGF-1, helix 3 adopts the 310

helix conformation exclusively (20). The helix 3 conformationof IGF-1 is intermittent between the other structures when incomplex with the IGFBP-5 fragment (26). The variationsin helix 3 conformation have a pronounced affect on the loca-tion of Leu54, a key residue in the interactions with everyligand. Changes in residue orientation upon binding mayaccount for the ability of a small peptide hormone to recognizesix binding proteins and two receptors. The dynamic charac-teristics that make IGF-1 difficult to characterize in solution

260 Deshayes

may also permit the hormone to efficiently adjust conforma-tion in order to recognize dissimilar receptors.

II.C. Vascular Endothelial Growth Factor

Peptide phage display has been used to generate compellingevidence that proteins contain discrete sites that dominatethe binding interactions. An important example of the conser-vation of binding sites is the study of complexes between vas-cular endothelial growth factor (VEGF) and four unrelatedligands. VEGF is a primary modulator of physiological angio-genesis that exists in solution as a homodimer and functionsby inducing dimerization of its tyrosine kinase receptors,fms-like tyrosine kinase-1 (Flt-1) and kinase-insert domaincontaining receptor (KDR) (28). The crystal structure ofVEGF in complex with the second Ig-like domain of Flt-1

Figure 4 Structure of the complex between the phage-derivedpeptide F1-1 and IGF-1. The peptide is shown as a worm.


(Flt-1D2) revealed symmetric binding sites at the poles of thedimeric VEGF receptor-binding domain (29) that are similarto the KDR-binding sites (30–33) revealed by alanine-scan-ning mutagenesis. The binding sites for antibodies that neu-tralize VEGF have also been shown to overlap with theKDR- and Flt-1-binding sites (31,34,35).

A library of disulfide-constrained peptides was sortedagainst VEGF and yielded three peptide classes that boundto VEGF and blocked KDR binding (16). Representatives oftwo of these peptide classes, v108 and v107, have subse-quently had their structures determined in complex withVEGF (15,36). The structures reveal that the binding sitesfor both peptides overlap significantly with each other andthe receptor- and antibody-binding sites. The contact betweenv108 and VEGF involves primarily main-chain-mediatedhydrogen bonds (Fig. 5A), while in contrast v107 makesextensive hydrophobic side-chain-mediated contacts (Fig. 5B).

Each peptide complex resembles another structurallycharacterized VEGF complex. The binding mode of v108 mostclosely resembles that of an antigen-binding fragment (Fab)from the neutralizing anti-VEGF antibody Avastin (Fig. 5C)(34,35). The interaction of v107 with VEGF is most similarto that observed for Flt-1D2 (Fig. 5D) (29).

Although the modes of interaction are similar, the disso-ciation constant (Kd) values of the phage-derived peptides [0.6and 2.2 mM for v107 and v108, respectively] (15,16) are signif-icantly weaker than the 2–10 nM reported for Flt-1D2 and 13and 0.11 nM reported for the Avastin Fab and its affinity-opti-mized version, respectively (15,29,35). The 19-residue peptidev107 buries 1167 A2 of hydrophobic binding surface, while Flt-1D2 uses 101 residues to bury 1672 A2 of surface area. All theVEGF residues that contact v107 in the peptide complex alsocontact Flt-1D2. The 20-residue peptide v108 buries a totalsurface area of 1350 A2 upon binding to VEGF, while thereceptor-blocking Fabs bury 1750–1800 A2. Up to 13 intermo-lecular hydrogen bonds are observed in the interface of thev108 complex, while 10–12 are found in the Fab complexes.All VEGF atoms involved in hydrogen bonds in the v108 com-plex also hydrogen bond to the Fabs.

262 Deshayes

That phage-derived peptides adopt VEGF-binding inter-actions similar to those observed for protein ligands is notlikely due to random chance; rather, it suggests that onlylimited regions of the VEGF surface support high-affinitybinding interactions. An illuminating comparison can bemade using data obtained from crystal structure analysis ofprotein–protein complexes. Examination of 32 protein dimerstructures revealed that the size of the protein–protein inter-faces ranged from 368 to 4761 A(37). A similar analysis ofprotease inhibitor complexes and antibody–protein antigencomplexes showed a range of 1600 � 350 A (38). VEGF-binding peptides present interaction surfaces on reducedscaffolds, but the buried surface areas in these interfaces(�1000 A2) fall within the observed norms for protein–proteininteractions. Thus, in order to effectively bind to VEGF, it isnecessary to utilize a major part of the optimal bindingregion.

Figure 5 Structures of the interfaces within complexes of VEGFwith (A) peptide v1O8, (B) peptide v1O7, (C) the neutralizing FabAvastin, and (D) Flt-1D2.


II.D. Immunoglobulin G

Another phage-based study that points to the existence ofoptimal binding sites is the probing of the binding surfaceof the crystallizable fragment of human immunoglobulin G(IgG-Fc). It has been shown that the same hinge region ofthe IgG-Fc interacts with four different proteins: protein A(39), protein G (40), neonatal factor (41), and rheumatoidfactor (Fig. 6) (42).

Amazingly, four proteins recognize the same bindingsurface using four drastically different folds. Is it possiblethat the proteins bind to that region due to functional consid-erations and not because the intrinsic properties of thesite make it optimal for binding? In order to answer thesequestions, a library of disulfide-constrained peptides waspanned against the IgG-Fc. A single peptide, Fc-III, wasselected and further optimized to produce a 13-residuepeptide that inhibits protein A binding with a Ki of 25 nM.Structural analysis revealed that although Fc-III adopts ab-hairpin conformation distinct from the four proteins that

Figure 6 Structures of the IgG-Fc in complex with (A) rheuma-toid factor, (B) neonatal factor domain, (C) domain C2 of proteinG, (D) domain B1 of protein A, and (E) the phage-derived peptideFc-III.

264 Deshayes

bind to the hinge region, the peptide also binds to the samelocation (Fig. 6E) (13). Since there is no functional biasinvolved in the selection of peptides displayed on phage, thisresult indicates that the Fc region of immunoglobulin G hasphysical properties that make it the most suitable binding sitefor protein ligands. Although Fc-III is much smaller than thenatural ligands, the area of interaction (650 A2) is close to thatof the natural ligands (�740 A2). It appears that theconsensus binding patch on the IgG-Fc is an area of approxi-mately 700 A2 that is more adaptive and hydrophobic thanother surfaces of the protein.

The advantages of using phage display to guide the re-engineering of proteins was demonstrated in a paper in whichone of the helices from the B region of protein A was removedin order to create a minimized binding domain (Fig. 7) (39).Although the binding interface remained intact, the affinitywent from a Kd of 10 nM for protein A to greater than 1 mMfor the minimized protein. Using results from phage librariesthat explored three regions of the two helix structures, it wasfound that 12 mutations dramatically increased the a-helicalstructure of the miniprotein, thereby preorganizing thebinding surface. In effect, the mutations discovered usingphage display carry out the same function as the thirdhelix and restore the affinity for the IgF-Fc (Kd¼ 43 nM). Insubsequent work, the preorganization of the binding surface

Figure 7 Phage-derived peptides that mimic the IgG-Fc-bindingsurface of protein A. (A) The original optimized structure and(B) the structure of the 34-residue peptide containing a stabilizingdisulfide bond, shown in dark gray.


was increased by adding a disulfide bond between the twoa-helices, which eventually yielded a 34-residue variant(Fig. 7B) that binds to the Fc with a ninefold increase in affi-nity (43). Thus, affinity maturation yielded a peptide thatcontains half the number of residues found in the nativedomain and yet mimics both its structure and function.

II.E. Factor VII

Many enzymes are regulated via binding to so-called exosites,sites distant from the active site. Inhibition via exosite bindingcan occur either by rearrangement of the enzyme into a lessactive conformation or by the physical blocking of the interac-tion between enzyme and substrate. An important applicationof phage display is shown in the work of Dennis et al., whichproduced a novel 20-residue peptide (E-76) that acts as anexosite enzyme regulator of factor VII activity (Fig. 8) (5).

Factor VII is a serine protease that is activated by tissuefactor upon vascular damage and serves as a key componentin the formation of fibrin clots in the coagulation cascade(44). The coagulation cascade is a cluster of serine proteasesthat control the balance between hemostasis, blood vesselrepair, and thrombosis—the formation of artery-blockingclots. There is evidence that selective inhibition of thecoagulation cascade will maintain the essential balance

Figure 8 Structure of E-76 (dark gray) bound to factor VII. Keypeptide side chains that make contact with factor VII are shownand labeled.

266 Deshayes

between hemostasis and thrombosis, and thereby avertfurther cardiovascular degeneration (44).

The sorting of the peptide libraries was carried outagainst factor VII bound to tissue factor, so that peptides thatcompete with tissue factor binding could not be selected. E-76binds tightly to factor VII (Kd¼ 8.5 nM) and inhibits thefactor VII-mediated activation of factor X, the next componentin the coagulation cascade, with an IC50 of 1 nM. No bindingwas detected between E-76 and other members of the coag-ulation cascade. E-76 contains a disulfide bond that maintainsa well-defined solution structure consisting of a distorted typeI reverse turn and a type I helix (Fig. 8). This structurehas been shown to be maintained upon binding to factorVII (5).

Structural and functional data indicate that E-76 makescontact with a 660 A2 site distant from both the enzymeactive site and the tissue factor-binding site. Mutagenesisdata suggest that four hydrophobic residues (Leu2, Trp11,Tyr12, and Phe15) contribute an inordinate amount to thebinding energy, indicating that the binding surface is madeup of hydrophobic contacts between the peptide and theprotein.

E-76 displays a mixed inhibitor mechanism, indicatingthat it does not directly compete with factor X for binding tofactor VII. This result differs from other known exosite inhibi-tors. For example, the dodecapeptide hirugen, derived fromthe leech protein hirudin, binds at the fibrinogen-bindingsite and acts as a competitive inhibitor, unlike E-76, whichdisplays a mixed inhibitor mechanism (45,46). It has beensuggested that the formation of a binding surface to accommo-date E-76 rearranges the protein hydrogen-binding networkof factor VII so that the ‘‘oxyanion hole’’ is disrupted, therebyremoving a crucial feature of the serine protease mechanism(5). It is possible that the E-76 epitope has no equivalent innature and the inhibition mechanism is unique to E-76. Onecan speculate that blocking the tissue factor-binding sitehas biased the selection against the tightest binders, andthe unique nature of E-76 is a product of the selection process.This result highlights an interesting application of phage


display; by varying the panning conditions, it may be possibleto obtain diverse binders to a single protein. For example,once the primary binding site is occupied, can other bindersbe identified that recognize other regions of the protein?How many times can this process be repeated?

II.F. Summary of Extracellular Protein–ProteinInteractions

The inability of phage display to identify small molecule-binding sites on extracellular proteins is not due to a lack ofpotential binding epitopes being presented to the target pro-tein. Increases in library diversity from 107 to greater than1011 individual members has not changed the types of pep-tides selected (1). It appears that many, if not most, extracel-lular proteins preferentially bind epitopes that resemble‘‘miniproteins.’’ There are exceptions (e.g., Eph recep-tor=ephrin (47), BAFF=Blys (3)) in which extracellular pro-teins have evolved binding sites for small epitopes, but themajority of extracellular proteins studied to date have givenresults similar to the VEGF example. Thus, we conclude thatif diverse peptide libraries exclusively produce large,extended binding surfaces, the protein is unlikely to be a goodtarget for small molecule therapeutics.

III. INTRACELLULAR PROTEIN–PROTEININTERACTIONS

The accumulated evidence suggests that the majority ofextracellular proteins contain widely dispersed binding sites.This generality does not appear to carry over to intracellularprotein–protein interactions. Small, continuous epitopescontained within large proteins are often recognized byintracellular receptors. For example, proline-rich sequencesare recognized by two different domains: Src homology 3(SH3) domains (48) and WW domains (named for two highlyconserved tryptophan residues found in the consensussequence) (49). Another example is phosphotyrosine, which

268 Deshayes

is recognized by both the Src homology 2 (SH2) (50) andphosphotyrosine-binding (PTB) domains (51).

III.A. PDZ Domains

A novel binding module, the PDZ domain (named after thefirst proteins in which it was observed: postsynaptic den-sity-95, discs large, and zonula occludens-1) that recognizesthe C-terminal sequences, has attracted a lot of attention(52). The PDZ domain is a compact module consisting ofapproximately 90 amino acids that are found imbedded in alarger protein, often with other PDZ modules or other moduletypes. Multiple modules can act in concert to bind and localizemultiple partners and construct intracellular architecture. Inmany cases, short peptides of 4–8 residues are specific, potentligands. The structure of a phage-optimized ligand bound tothe PDZ domain of the protein Erbin is shown in Fig. 9 and

Figure 9 Structure of a phage-optimized peptide (WETWV)bound to the Erbin PDZ domain. The C-terminus of the peptide islabeled and peptide side chains are shown.


is representative of how these receptors recognize specific C-termini (53). The peptide main chain has antiparallel b-sheetinteractions with the PDZ domain main chain and the term-inal carboxylate is inserted into a ‘‘carboxylate-binding loop.’’Essentially all PDZ domains maintain these interactions.

The unique nature of PDZ domain-mediated recognitionmakes it possible to use phage display to explore ligand speci-ficity of domain families, understand the structure–functionrelationships, and predict and validate putative natural pro-tein–protein interactions. PDZ domains assemble proteinsinto functional complexes localized at specific subcellular siteswithin eukaryotic cells: epithelial tight junctions or neuronalsynaptic densities, for example (54,55). Novel C-terminallydisplayed peptide-phage libraries were used to investigatethe binding specificities of two of the six PDZ domains froma membrane-associated guanylate kinase (MAGI-3) (56). Eachdomain bound specifically to small peptides and the binding spe-cificities were very different from each other, suggesting thatthe two PDZ domains bind different ligands. Like many PDZ-containing proteins, MAGI-3 contains multiple PDZ domains,and thus, a single MAGI-3 protein likely acts as a multiligandscaffold to assemble multicomponent protein complexes.

More recently, C-terminal peptide-phage libraries wereused to study the binding specificity of the single PDZ domainof Erbin (57), a protein that was originally identified as aputative ligand for ErbB2 in a yeast-two-hybrid screen (58).ErbB2 is an epidermal growth factor receptor-related tyrosinekinase that is a causal factor in the development of somecancers (59). The intent was to discover high-affinity ligandsfor the Erbin PDZ domain that could be used to disrupt itsinteraction with ErbB2. Surprisingly, phage display revealeda binding consensus for Erbin PDZ ([D=E][T=S]WVCOOH) thatdiffered significantly from the C-terminal sequence of ErbB2(DVPV

cooh) (56). Furthermore, searches of genomic databases

revealed that the phage-derived consensus closely matchedthe C-termini of d-catenin and two related homologs (ARVCFand p0071), which all terminate in an identical sequence(DSWVCOOH). Since these catenins are also mediators ofintracellular signaling (60,61), it is possible that the

270 Deshayes

interaction with Erbin is physiologically relevant. Subsequentin vitro and in vivo experiments clearly demonstrated thatErbin binds to d-catenin and its homologs with high affinityand specificity, while its affinity for ErbB2 is significantlylower. Subsequent work demonstrated that neither Erbinnor the C-terminus of ErbB2 interacts in the proposed man-ner (62). Furthermore, the in vivo interaction between Erbinand ARVCF was successfully disrupted by intracellular deliv-ery of phage-derived high-affinity peptides (57), thus demon-strating the utility of peptide ligands for intracellular targetvalidation.

In studying the relationships between PDZ domain struc-ture and function, we have made extensive use of invitro affinity assays with synthetic peptides to accuratelymap the determinants of affinity and specificity. Our resultssuggest that PDZ domains can use up to five side chains atthe C-termini of proteins to bind with high affinity to their cog-nate ligands while excluding other closely related sequences.This point was illustrated by comparing the binding specificityof the Erbin PDZ domain to that of MAGI-3 PDZ2. Peptide-phage libraries revealed that these two domains recognizeC-terminal consensus sequences that differ at only one sitein the last four positions ([D=E][T=S]WVCOOH vs. [C=V=I][T=S]WVCOOH for Erbin PDZ and MAGI-3 PDZ2, respectively).However, this single difference was sufficient to alter affinityby at least two orders of magnitude; a peptide bearing a gluta-mate side chain (TGWETWVCOOH) interacted exclusively withthe Erbin PDZ domain, while a peptide in which glutamate wasreplaced with isoleucine (TGWITWVCOOH) interacted onlywith MAGI-3 PDZ2 (57).

The type of ligand side chains accepted by a PDZ domaindepends on the binding surface defined by the side chainresidues that line the peptide-binding groove. Analysis ofthese interactions with phage-derived peptides gives deepinsights into the manner in which PDZ domains recruit speci-fic targets (53). Furthermore, the observation that PDZdomains bind short linear peptides with high affinity andspecificity suggests that they may be valid small moleculetargets.


III.B. Inhibitors of Apoptosis

A second class of receptor proteins that recognize short linearamino acid epitopes is the inhibitor of apoptosis proteins(IAPs), which provide protection for cells against diversepro-apoptotic stimuli (63). The IAPs were first identified inthe baculovirus, where they suppress the cell death responseduring viral infection (64,65), and they were subsequentlydetected in both invertebrates and vertebrates (58,66–74).The IAP family is characterized by the presence of thebaculovirus IAP repeat (BIR) motif. BIRs are �70-residuezinc-binding domains that bind to, and thereby inhibit, thecaspase proteases that mediate apoptosis.

The ubiquitously expressed X-chromosome-linked IAP(X-IAP) contains three BIR domains; the second BIR domain(BIR2), together with the immediately preceding linkerregion, inhibits active caspase-3 and caspase-7 (75–79), whilethe third BIR domain (BIR3) is a specific inhibitor of caspase-9activation (77,80,81). Another interesting member of the IAPfamily is melanoma IAP (ML-IAP), a protein that is not detect-able in most normal adult tissues but is strongly overex-pressed in melanoma cells (82,83). ML-IAP contains a singleBIR domain and has also been shown to be a strong inhibitorof apoptosis (58,82–84). Caspase-9 binds to BIR3 of X-IAP lar-gely through interactions involving the N-terminus of thesmall subunit of caspase-9 (AVPT) (84–86). Importantly, theexposed N-terminus of the small subunit of caspase-9 is homo-logous to the processed N-termini of natural antagonists of theIAPs, including the mammalian proteins Smac=DIABLO(AVPI) and HtrA2=Omi (AVPF). IAP antagonists promoteapoptosis by releasing caspases from the BIR domains andthereby allowing the apoptotic cascade to proceed (63).

Phage display was used to investigate the sequence diver-sities that bind to BIR domains. Peptide libraries were sortedagainst two BIR domains from X-IAP and the BIR domainfrom ML-IAP (87). Only the four extreme N-terminal positionsshow sequence consensus, indicating that only these positionsare essential for recognition by the BIR domain. Significantobservations are that alanine was exclusively selected at the

272 Deshayes

N-terminus, and proline is highly favored in the third positionfor X-IAP BIR3 and ML-IAP BIR, but not for BIR2. Anotherdifference is the preference for glutamate in the second posi-tion of X-IAP BIR2 ligands, whereas the other BIR domainsprefer small hydrophobic residues such as valine or isoleucine.The last difference is that ML-IAP and X-IAP BIR3 select aro-matic amino acids in position 4, while X-IAP BIR2 preferssmall hydrophobic residues in this position.

Several peptide sequences derived from the phageresults were synthesized and assayed against three BIRdomains with fascinating results; both ML-IAP and X-IAPBIR2 can tolerate an acidic group in the second position, whileML-IAP BIR cannot (87). A combination of x-ray crystal struc-ture analysis and homology modeling was used to explain thedifferences in selectivity between X-IAP BIR3 and ML-IAPBIR (Fig. 10).

X-IAP BIR3 has an aspartate residue proximal to theposition 2 binding site, which precludes the placement of a

Figure 10 Structures of a peptide (AVPI) bound to (A) the BIR3domain of X-IAP and (B) the BIR domain of ML-IAP. Note thetwo key differences in sequence that contribute to differences inspecificity; Asp and Tyr residues in X-IAP BIR3 are substitutedby Ser and Phe, respectively, in ML-IAP BIR. Peptide side chainsare shown and labeled. Side chains on the BIR domains are labeledin italics.


negatively charged residue in this site (Fig. 10A). The aspar-tate is replaced by a serine in ML-IAP BIR and this residue iscapable of hydrogen bonding to a carboxylate group residingat position 2 of the ligand (Fig. 10B). Another interestingobservation is that ML-IAP BIR can more readily tolerateb-branched amino acids such as valine in position 3 thanX-IAP BIR3. This difference is explained by the change of aphenylalanine in ML-IAP BIR to the more sterically demand-ing tyrosine in X-IAP BIR3.

The final position for selectivity pointed out by thephage display experiment is that both ML-IAP BIR and X-IAP BIR3 have deep, adaptive, binding pockets at position4, and can therefore accommodate large aromatic aminoacids. In contrast, position 4 of X-IAP BIR2 is bettersuited for small hydrophobic amino acids such as valineand isoleucine.

The exceptional feature of this application of peptidephage display is that one experiment reveals the roots ofspecificity between three closely related domains. Subse-quently, binding data were combined with structural analysisto construct a detailed model of how BIR domains recognizecontiguous linear epitopes. It is difficult to imagine anothermethodology that can so rapidly elucidate so many keyfeatures of a binding interface.

IV. CONCLUSIONS

This chapter has highlighted the utility of phage-displayedpeptide libraries for investigating protein–protein interac-tions. The advances in our ability to display peptide librarieson phage are accelerating in concert with the exploration ofprotein–protein interactions. These gains in technology willhelp us more efficiently harvest the information generatedby the ongoing genomics and proteomics efforts. Phagedisplay not only enables the rapid analysis of many of thethousands of proposed protein–protein interactions, but alsoallows us to assess the potential of using small molecules tocontrol these interactions.

274 Deshayes

REFERENCES

1. Sidhu SS, Fairbrother WJ, Deshayes K. Exploring protein–protein interactions with phage display. Chem Bio Chem2003; 4(1):14–25.

2. Cochran AG. Antagonists of protein–protein interactions.Chem Biol 2000; 9:R85–R94.

3. Gordon NC, et al. BAFF=BLyS receptor 3 comprises a minimalTNF receptor-like module that encodes a highly focusedligand-binding site. Biochemistry 2003; 42(20):5977–5983.

4. Lowman HB, et al. Molecular mimics of insulin-like growthfactor 1 (IGF-1) for inhibiting IGF-1: IGF-binding proteininteractions. Biochemistry 1998; 37(25):8870–8878.

5. Dennis MS, et al. Peptide exosite inhibitors of factor VIIa asanticoagulants. Nature 2000; 404(6777):465–470.

6. Nakamura GR, et al. A novel family of hairpin peptides thatinhibit IgE activity by binding to the high-affinity IgE recep-tor. Biochemistry 2001; 40(33):9828–9835.

7. Nakamura GR, et al. Stable ‘‘zeta’’ peptides that act as potentantagonists of the high-affinity IgE receptor. Proc Natl AcadSci USA 2002; 99:1303–1308.

8. Skelton NJ, et al. Amino acid determinants of b-hairpin confor-mation in erythropoietin receptor agonist peptides derivedfrom a phage display library. J Mol Biol 2002; 316:1111–1125.

9. Deshayes K, et al. Rapid identification of small binding motifswith high-throughput phage display: discovery of peptidicantagonists of IGF-1 function. Chem Biol 2002; 9:495–505.

10. Livnah O, et al. Functional mimicry of a protein hormone by apeptide agonist: the EPO receptor complex at 2.8 A. Science1996; 273(5274):464–471.

11. Livnah O, et al. An antagonist peptide–EPO receptor complexsuggests that receptor dimerization is not sufficient for activa-tion. Nat Struct Biol 1998; 5:993–1003.

12. Eckert DM, et al. Inhibiting HIV-1 entry: discovery ofD-peptide inhibitors that target the gp41 coiled-coil pocket.Cell 1999; 99(1):103–115.


13. DeLano WL, et al. Convergent solutions to the binding at aprotein–protein interface. Science 2000; 287:1279–1283.

14. Scherf T, et al. A b-hairpin structure in a 13-mer peptide thatbinds a-bungarotoxin with high affinity and neutralizes itstoxicity. Proc Natl Acad Sci USA 2001; 98:6629–6634.

15. Pan B, et al. Solution structure of a phage-derived peptideantagonist in complex with vascular endothelial growth factor.J Mol Biol 2002; 316:769–787.

16. Fairbrother WJ, et al. Novel peptides selected to bind vascularendothelial growth factor target the receptor-binding site.Biochemistry 1998; 37(51):17754–17764.

17. de Vos AM, Ultsch MH, Kossiakoff AA. Human growthhormone and extracellular domain of its receptor: crystalstructure of the complex. Science 1992; 255:306–312.

18. Livnah O, et al. Crystallographic evidence for preformeddimers of erythropoietin receptor before ligand activation.Science 1999; 283(5404):987–990.

19. Remy I, Wilson IA, Michnick SW. Erythropoietin receptoractivation by a ligand-induced conformation change. Science1999; 283(5404):990–993.

20. Schaffer ML, et al. Complex with a phage display-derivedpeptide provides insight into the function of insulin-likegrowth factor I. Biochemistry 2003; 42:9324–9334.

21. De Meyts P. The structural basis of insulin and insulin-likegrowth factor-I receptor binding and negative co-operativity,and its relevance to mitogenic versus metabolic signalling.Diabetologia 1994; 37(suppl 2):S135–S148.

22. Ballard FJ, et al. Effects of interactions between IGFBPs andIGFs on the plasma clearance and in vivo biological activitiesof IGFs and IGF analogs. Growth Regul 1993; 3(1):40–44.

23. Sato A, et al. Three-dimensional structure of human insulin-like growth factor-I (IGF-I) determined by 1H-NMR anddistance geometry. Int J Peptide Protein Res 1993; 41(5):433–440.

24. Cooke RM, Harvey TS, Campbell ID. Solution structure ofhuman insulin-like growth factor 1: a nuclear magnetic

276 Deshayes

resonance and restrained molecular dynamics study. Biochem-istry 1991; 30(22):5484–5491.

25. Vajdos FF, et al. Comprehensive functional maps of the anti-gen-binding site of an anti-ErbB2 antibody obtained withshotgun scanning mutagenesis. J Mol Biol 2002; 320:415–428.

26. Zeslawski W, et al. The interaction of insulin-like growthfactor-I with the N-terminal domain of IGFBP-5. EMBO J2001; 20(14):3638–3644.

27. Vajdos FF, et al. Crystal structure of human insulin-likegrowth factor-1: detergent binding inhibits binding proteininteractions. Biochemistry 2001; 40(37):11022–11029.

28. Folkman J. Angiogenesis in cancer, vascular, rheumatoid andother disease. Nat Med 1995; 1:27–31.

29. Wiesmann C, et al. Crystal structure at 1.7 A resolution ofVEGF in complex with domain 2 of the Flt-1 receptor. Cell1997; 91(5):695–704.

30. Keyt BA, et al. The carboxyl-terminal domain (111–165) ofvascular endothelial growth factor is critical for its mitogenicpotency. J Biol Chem 1996; 271(13):7788–7795.

31. Muller YA, et al. Vascular endothelial growth factor: crystalstructure and functional mapping of the kinase domain recep-tor binding site. Proc Natl Acad Sci USA 1997; 94(14):7192–7197.

32. Fuh G, et al. Requirements for binding and signaling of thekinase domain receptor for vascular endothelial growth factor.J Biol Chem 1998; 273(18):11197–11204.

33. Li B, et al. Receptor-selective variants of human vascularendothelial growth factor. Generation and characterization.J Biol Chem 2000; 275(38):29823–29828.

34. Muller YA, et al. VEGF and the Fab fragment of a humanizedneutralizing antibody: crystal structure of the complex at 2.4 Aresolution and mutational analysis of the interface. Structure1998; 6(9):1153–1167.

35. Chen Y, et al. Selection and analysis of an optimized anti-VEGF antibody: crystal structure of an affinity-matured Fabin complex with antigen. J Mol Biol 1999; 293(4):865–881.


36. Wiesmann C, et al. Crystal structure of the complex betweenVEGF and a receptor-blocking peptide. Biochemistry 1998;37(51):17765–17772.

37. Jones S, Thornton JM. Protein–protein interactions: a reviewof protein dimer structures. Prog Biophys Mol Biol 1995;63:31–65.

38. Davies DR, Padlan EA, Sheriff S. Antibody–antigen com-plexes. Annu Rev Biochem 1990; 59:439–473.

39. Deisenhofer J. Crystallographic refinement and atomic modelsof a human Fc fragment and its complex with fragment B ofprotein A from Staphylococcus aureus at 2.9- and 2.8-A resolu-tion. Biochemistry 1981; 20(9):2361–2370.

40. Sauer-Eriksson AE, et al. Crystal structure of the C2 fragmentof streptococcal protein G in complex with the Fc domain ofhuman IgG. Structure 1995; 3(3):265–278.

41. Burmeister WP, Huber AH, Bjorkman PJ. Crystal structure ofthe complex of rat neonatal Fc receptor with Fc. Nature 1994;372(6504):379–383.

42. Corper AL, et al. Structure of human IgM rheumatoid factorFab bound to its autoantigen IgG Fc reveals a novel topologyof antibody–antigen interaction. Nat Struct Biol 1997; 4(5):374–381.

43. Starovasnik MA, Braisted AC, Wells JA. Structural mimicry ofa native protein by a minimized binding domain. Proc NatlAcad Sci USA 1997; 94(19):10080–10085.

44. Nemerson Y. Tissue factor and hemostasis. Blood 1988;71(1):1–8.

45. Skrzypczak-Jankun Eea. Structure of hirugen and hirilog 1complexes of alpha-thrombin. J Mol Biol 1991; 221:1379–1393.

46. Stubbs MT, Bode WA. A player of many parts: the spotlightfalls on thrombin’s structure. Thrombin Res 1993; 69:1–58.

47. Himanen JP, et al. Crystal structure of an Eph receptor–ephrin complex. Nature 2001; 414(6866):933–938.

48. Ren R, et al. Identification of a ten-amino acid proline-richSH3 binding site. Science 1993; 259:1157–1161.

278 Deshayes

49. Chen HI, Sudol M. The WW domain of Yes-associated proteinbinds a proline-rich ligand that differs from the consensusestablished for Src homology 3-binding modules. Proc NatlAcad Sci USA 1995; 92:7819–7823.

50. Songyang Z, et al. SH2 domains recognize specific phosphopep-tide sequences. Cell 1993; 72:767–778.

51. Kavanaugh WM, Williams LT. An alternative to SH2 domainsfor binding tyrosine-phosphorylated proteins. Science 1994;266:1862–1865.

52. Cowburn D. Peptide recognition by PTB and PDZ domains.Curr Opin Struct Biol 1997; 7:835–838.

53. Skelton NJ, et al. Origins of PDZ domain ligand specificity.Structure determination and mutagenesis of the Erbin PDZdomain. J Biol Chem 2003; 278(9):7645–7654.

54. Tomita S, Nicoll RA, Bredt DS. PDZ protein interactions reg-ulating glutamate receptor function and plasticity. J Cell Biol2001; 153:F19–F23.

55. Fanning AS, Anderson JM. Protein modules as organizersof membrane structure. Curr Opin Cell Biol 1999; 11:432–439.

56. Fuh G, et al. Analysis of PDZ domain–ligand interactionsusing carboxyl-terminal phage display. J Biol Chem 2000;275:21486–21491.

57. Laura RP, et al. The Erbin PDZ domain binds with high affi-nity and specificity to the carboxyl termini of d-catenin andARVCF. J Biol Chem 2002; 277:12906–12914.

58. Borg J-P, et al. ERBIN: a basolateral PDZ protein that inter-acts with the mammalian ERBB2=HER2 receptor. Nat CellBiol 2000; 2:407–414.

59. Klapper LN, et al. Biochemical and clinical implications ofthe ErbB=HER signaling network of growth factor receptors.Adv Cancer Res 2000; 77:25–79.

60. Lu Q, et al. Brain armadillo protein delta-catenin interactswith Abl tyrosine kinase and modulates cellular morphogen-esis in response to growth factors. J Neurosci Res 2002; 67:618–624.


61. Fraser PE, et al. Presenelin function: connections to Alzhei-mer’s disease and signal transduction. Biochem Soc Symp2001; 67:89–100.

62. Shelly M, et al. Polar expression of ErbB-2=HER2 in epithelia:bimodal regulation by Lin-7. Dev Cell 2003; 5:475–486.

63. Salvesen GS, Duckett CS. IAP proteins: blocking the road todeath’s door. Nat Rev Mol Cell Biol 2002; 3(6):401–410.

64. Crook NE, Clem RJ, Miller LK. An apoptosis-inhibiting bacu-lovirus gene with a zinc finger-like motif. J Virol 1993; 67(4):2168–2174.

65. Birnbaum MJ, Clem RJ, Miller LK. An apoptosis-inhibitinggene from a nuclear polyhedrosis virus encoding a polypeptidewith Cys=His sequence motifs. J Virol 1994; 68(4):2521–2528.

66. Rothe M, et al. The TNFR2–TRAF signaling complex containstwo novel proteins related to baculoviral inhibitor of apoptosisproteins. Cell 1995; 83(7):1243–1252.

67. Roy N, et al. The gene for neuronal apoptosis inhibitory pro-tein is partially deleted in individuals with spinal muscularatrophy. Cell 1995; 80(1):167–178.

68. Hay BA, Wassarman DA, Rubin GM. Drosophila homologs ofbaculovirus inhibitor of apoptosis proteins function to blockcell death. Cell 1995; 33(7):1253–1262.

69. Duckett CS, Nava VE, Gedrich RW, Clem RJ, Van Dongen JL,Gilfillan MC, Shiels H, Hardwick JM, Thompson CB.A conserved family of cellular genes related to the baculovirusiap gene and encoding apoptosis inhibitors. EMBO J 1996;15(11):2685–2694.

70. Liston P, et al. Suppression of apoptosis in mammalian cells byNAIP and a related family of IAP genes. Nature 1996;379(6563):349–353.

71. Uren AG, et al. Cloning and expression of apoptosis inhibitoryprotein homologs that function to inhibit apoptosis and=orbind tumor necrosis factor receptor-associated factors. ProcNatl Acad Sci USA 1996; 93(10):4974–4978.

72. Ambrosini G, Adida C, Altieri DC. Activation-dependentexposure of the inter-EGF sequence Leu83–Leu88 in factor Xa

280 Deshayes

mediates ligand binding to effector cell protease receptor-1.J Biol Chem 1997; 272(13):8340–8345.

73. Fraser AG, et al. Caenorhabditis elegans inhibitor of apoptosisprotein (IAP) homologue BIR-1 plays a conserved role incytokinesis. Curr Biol 1999; 9(6):292–301.

74. Uren AG, et al. Role for yeast inhibitor of apoptosis (IAP)-likeproteins in cell division. Proc Natl Acad Sci USA 1999;96(18):10170–10175.

75. Takahashi R, et al. A single BIR domain of XIAP sufficientfor inhibiting caspases. J Biol Chem 1998; 273(14):7787–7790.

76. Sun C, et al. NMR structure and mutagenesis of the inhi-bitor-of-apoptosis protein XIAP. Nature 1999; 401(6755):818–822.

77. Chai J, et al. Structural basis of caspase-7 inhibition by XIAP.Cell 2001; 104(5):769–780.

78. Huang Y, et al. Structural basis of caspase inhibition by XIAP:differential roles of the linker versus the BIR domain. Cell2001; 104(5):781–790.

79. Riedl SJ, et al. Structural basis for the inhibition of caspase-3by XIAP. Cell 2001; 104(5):791–800.

80. Deveraux QL, et al. Cleavage of human inhibitor of apoptosisprotein XIAP results in fragments with distinct specificities forcaspases. EMBO J 1999; 18(19):5242–51.

81. Sun C, et al. NMR structure and mutagenesis of the third Birdomain of the inhibitor of apoptosis protein XIAP. J Biol Chem2000; 275(43):33777–33781.

82. Vucic D, et al. ML-IAP, a novel inhibitor of apoptosis that ispreferentially expressed in human melanomas. Curr Biol2000; 10(12):1359–1366.

83. Kasof GM, Gomes BC. Livin, a novel inhibitor of apoptosisprotein family member. J Biol Chem 2001; 276(5):3238–3246.

84. Vucic D, et al. SMAC negatively regulates the anti-apoptoticactivity of melanoma inhibitor of apoptosis (ML-IAP). J BiolChem 2002; 277(14):12275–12279.


85. Srinivasula SM, et al. A conserved XIAP-interaction motif incaspase-9 and Smac=DIABLO regulates caspase activity andapoptosis. Nature 2001; 410(6824):112–116.

86. Shiozaki EN, et al. Mechanism of XIAP-mediated inhibition ofcaspase-9. Mol Cell 2003; 11(2):519–527.

87. Franklin MC, et al. Structure and function analysis of peptideantagonists of melanoma inhibitor of apoptosis (ML-IAP).Biochemistry 2003; 42(27):8223–8231.

282 Deshayes

7

Substrate Phage Display

SHUICHI OHKUBO

Cancer Research Laboratory, Hanno ResearchCenter, TAIHO Pharmaceutical Co., Ltd., Hanno,

Saitama, Japan

I. OVERVIEW

‘‘Substrate phage display’’ is a powerful application thatmakes it possible to screen substrate sequences of enzymesfrom large and diverse collections of randomized sequenceswithout any initial substrate data. The sensitivity and versa-tility of this technique have been clearly established throughthe discrimination of the substrate specificity of closelyrelated proteases or protein tyrosine kinases. The informationobtained from substrate phage display has substantiallyimproved our understanding of substrate recognition in cata-lysis and signal transduction. This chapter highlights recentadvances in the field of substrate phage display and illus-trates its utility in cancer research.

283

II. INTRODUCTION

Dysregulation of specific enzymes, such as proteases, proteinkinases, and protein phosphatases, has been implicated invarious pathological conditions, especially malignancies.Therefore, much effort has been directed towards developingpotent inhibitors of specific enzymes for cancer chemopreven-tion. However, many of these compounds are broad-spectruminhibitors that show undesirable side effects among theseclosely related enzymes. Thus, there is still a considerableneed for more selective inhibitors and for new and high-qual-ity treatments. A better understanding of substrate specifici-ties of these enzymes may significantly improve our overallknowledge about these enzymes, and this will facilitate thedesign and optimization of potent and selective inhibitors.In recent years, high-throughput screening (HTS), wherehundreds of thousands of compounds can be tested for activityduring a short period, has been increasingly used to discovernovel lead candidate molecules. A key step in establishing anHTS format is to identify a highly selective substrate for usein enzyme assays. A basic understanding of substrate prefer-ences allows further clarification of the physiological roles ofthese enzymes.

Traditionally, the substrate specificities of enzymes havebeen studied using synthetic peptides corresponding tosequences derived from known substrate proteins. However,in the cases of proteases and protein tyrosine kinases it hasbeen shown that the sequences derived from natural sub-strates are not always optimal for these enzymes when thecatalytic activities are studied in vitro with synthetic peptides(1–4). Moreover, this approach does not permit the identifica-tion of novel substrate sequences and thereby new targetmolecules, which remain largely unknown.

‘‘Substrate phage display’’ is a powerful application ofthe phage display technique, which enables us to screen sub-strate sequences of enzymes from large, diverse collections ofrandomized sequences without any initial substrate data(5–7). In contrast to other combinatorial approaches, sub-strate phage display has an added advantage, in that it

284 Ohkubo

can generate a vast number of possible combinations,thereby enabling rapid library construction and substrateoptimization in a cost-effective manner (5,6,8–10). Substratephage display can be used to identify the substrate seq-uences of proteases (5), protein serine=threonine kinases (11),and protein tyrosine kinases (10). In this chapter, we sum-marize the basic concepts of substrate phage display anddiscuss how they might be used to facilitate cancer research.

III. THE CONCEPT OF SUBSTRATE PHAGEDISPLAY

The concept of substrate phage display has been aptlydescribed by Gram (7), as follows: ‘‘In general terms, thesubstrate phage concept is based upon the principle that aphage library of potential substrate peptides is subjected tomodification by an enzyme, followed by selection of thosephage displaying a modified or catalytically processed pep-tide.’’ Substrate phage display was first introduced by Mat-thews and Wells (5) to identify substrates for variousproteases, including subtilisin BPN0 mutant, factor Xa, andHIV protease. Since then, this technique has been success-fully used to identify substrates for several types of proteasesand protein kinases, as described in more detail below.

III.A. Screen of Substrate Sequencesfor Proteases

Substrate phage display has been used to identify substrates forseveral classes of endopeptidases, including serine-, aspartyl-,and metallo-proteases. Table 1 summarizes the various pro-tease substrates and their cleavage sites that have been identi-fied by substrate phage display to date (2–5,9,12–30).

Two strategies, monovalent and multivalent display,have been used for the selection and optimization of substratesfor proteolytic enzymes (5,9). In either case, gene III of fila-mentous phage is modified such that a randomized sequenceis linked to the N-terminus of protein III (pIII) and an

Substrate Phage Display 285

Table 1 Substrate Preferences of Proteases Identified Using Substrate Phage Display

Protease Substrate sequence Reference

Serine protease Subtilisin BPN0 mutantS24C=H64A=E156S=G166A=G169A=Y217L

TSM#HT (5)

Subtilisin BPN0 mutant N62D=G166D GNLMRK#G (12)Human factor Xa (G=A=T=F)R# (5)Mouse furin (L=P)RRF(K=R)#RP (13)Human tissue-type plasminogen GGSGPFGR#SALVPEE (2)

activator (t-PA) FRGR#Ka (14)Human urokinase-type plasminogen

activator (u-PA)GSGK#Sb (15)

HSV-1 protease LVLA#SSSF (16)Mouse tryptase SLSSR#QSP (17)Rat granzyme B IEXD#XG (18)Human prostate specific antigen (PSA) SS(Y=F)Y#S(G=S) (3)Human elastase mutant H57A MEHV#VY (19)Human plasmin GIYR#SR (4)Human membrane-type serine protease 1 (R=K)XSR#A (20)

X(R=K)SR#AHuman a-thrombin PR#G (21)

GR#R#G

Human kallikrein 2 (hK2) LR#SRA (22)Rat mast cell protease 4(rMCP-4) LVWF#RG (23)Staphylococcus aureus signal peptidase SpsB LPASLPSF (24)

286

Ohku

bo

Aspartyl HIV protease SQNYPIVQ (5)protease GSGIF#LETSL (25)

Metalloprotease Humangelatinase A(MMP-2) PXX#XHyc (26)

(L=I)XX#XHy

XHySX#LHXX#XHy

Stromelysin 1 (MMP-3) PFE#LRA (9)Matrilysin (MMP-7) PLE#LRA (9)Human gelatinase B (MMP-9) PR(S=T)#XHy(S=T) (27)Human collagenase 3 (MMP-13) GPLG#MRGL (28)Human menbrane type-1 matrix PX(G=P)#L (29)

metalloproteinase (MT1-MMP, MMP-14) RIGF#LRTAd (30)

#site of digestion.aMinimized, t-PA selective sequence.bMinimized, u-PA selective sequence.cXHy is a hydrophobic residue.dOne of the highly selective sequence.

Substrate

Phage

Disp

lay287

additional affinity tag sequence that enables attachment of thephage to an immobile phase is linked N-terminal to the ran-dom sequence (Fig. 1). As shown in Fig. 2, this phage libraryis incubated with a protease and uncleaved phage can beexcluded from the solution by affinity binding. Uncapturedphage carry a cleavable substrate site within the randomizedsequence that can easily be recovered and subjected torepeated selection and amplification cycles. In another app-roach, the phage library is first allowed to bind to a solid sup-port via the affinity sequence. The phage are then subjected toproteolysis to release those phage that express peptidesequences that are susceptible to the enzyme. The releasedphage are then amplified, bound, and cleaved again.

Protein or peptide epitopes, or histidine or FLAG tagshave been successfully used as affinity tags to identify pro-tease substrate sequences (5,9,16,18,27). In order to reducethe frequency of false positive clones during the selection,a high binding affinity between the tag and its correspond-ing matrix is desired. For example, a phage that includeda histidine tag sequence at the N-terminus of pIII and exhi-bited high-affinity binding for nickel-nitrilotriacetic acid(Ni-NTA), with an equilibrium binding constant (KD) of12 nM (21).

Figure 1 Schematic representation of phage particles used to con-struct a substrate phage library to identify substrates for proteases.For simplicity, this schematic representation of the phage particleshows only one pIII on the phage body. The substrate phage libraryconsists of the affinity tag sequence and the protease target rando-mized sequence at the N-termini of pIII of filamentous phage.

288 Ohkubo

III.B. Screen of Substrate Sequences for ProteinKinases

As described in more detail below, Westendorf et al. (11) useda different phage display approach to identify substrates ofprotein kinases. These investigators isolated phage clonesthat can be phosphorylated by partially purified protein seri-ne=threonine kinases. This approach has also been used todetermine the substrate specificity of protein tyrosine kinases(10). Table 2 provides a summary of the various kinasesubstrates and their phosphorylation sites that have beenidentified to date (10,11,31,32).

The insertion of randomized sequences into pIII offilamentous phage is the most common approach for making

Figure 2 Schematic illustration of the use of substrate phageselection to select substrates for proteases. (A) Phage displayingrandomized substrates and an affinity tag fused to pIII areincubated with the protease of interest. (B) The uncleaved (nonsub-strate) phage is captured by tag-binding materials. (C) The phagedisplaying substrates susceptible to protease cleavage are recoveredand amplified in Escherichia coli. (D) Individual clones are eithersubjected to DNA sequencing or the amplified phage is used for asubsequent round of selection.


Table 2 Substrate Preferences of Protein Kinases Identified Using Substrate Phage Display

Kinase Substrate sequence Reference

Serine=threonine kinase Human partial purified M-phasekinasesa

LTPLKb (11)

Tyrosine kinase Bovine Fyn E(XHy=T) Y GXXHyc (31)

Human c-Src (D=E)X(I=L)Y(G=W)X(F=W)X

(10)

Human Blk XXI Y (D=E)XLP (10)Human Lyn (D=E)X(I=L) Y (D=E)XLP (10)Human Syk XXD Y EXXX (10)Human Tie-2 RLVA Y EGWV (32)

T: phosphorylated threonine residue; Y: Phosphorylated tyrosine residue.aPartial purified kinase-containing fraction prepared from M-phase arrested HeLa cells.bSelected by MPM2 antibody, which can recognize a phospho amino acid-containing epitope of M-phase phosphoproteins.cXHy is a hydrophobic residue.

290

Ohku

bo

phage display libraries (6). In this screen, gene III is also mod-ified such that a randomized sequence is expressed at theN-terminus of pIII. A schematic overview of a selection forkinase substrates is shown in Fig. 3. To identify and charac-terize the substrate sequences of protein kinases, a phagelibrary is incubated in the presence of protein kinases. Phagesexpressing peptides that can be efficiently phosphorylated areselected by incubation with an antibody specific for epitopescontaining a certain phosphoamino acid (11) or with antipho-sphotyrosine antibodies (10,31,32), followed by multiplerounds of conventional selection and amplification. After thefinal selection, individual clones are isolated and the sub-strate peptide sequences are determined by sequencing therelevant portion of the phage DNA.

Figure 3 Schematic illustration of the use of substrate phage selec-tion to select substrates for protein kinases. (A) Phage displaying ran-domized substrates are incubated with the protein kinase of interest.(B) The phosphorylated phage is captured by an antiphosphoaminoacid antibody, such as antiphosphotyrosine. (C) The phage displayingspecific substrates are recovered and amplified in Escherichia coli.(D) Individual clones are either subjected to DNA sequencing or theamplified phage is used for a subsequent round of selection.


III.C. Characterization of Proteolysis Reactionby Analysis of Substrate Phage

Matthews et al. (13) developed a method to identify the truepositive clones among the selected phage and to compare therelative rates at which the isolated sequences are hydrolyzed.Fusion proteins were prepared that contained an affinity pep-tide tag sequence, the substrate sequence obtained from thephage screen, and an alkaline phosphatase. These fusion pro-teins were then immobilized onto the affinity tag-bindingprotein and the time course of substrate hydrolysis wasfollowed by monitoring the activity of alkaline phosphatasereleased after incubation with protease. In a similar procedure,Cloutier et al. (22) used cyan fluorescent protein (CFP), a var-iant of the green fluorescent protein, in place of alkaline phos-phatase, and the time course of substrate hydrolysis wasfollowed by monitoring the released fluorescence.

Smith et al. (9) developed another simple and rapidmethod to compare relative rates of hydrolysis of the isolatedsequences. Isolated phage clones were incubated in a solutioncontaining the protease of interest and the mixture was thenspotted onto nitrocellulose membranes. The membranes werethen probed with an antibody directed against the affinitytag. If the tag had been hydrolytically removed from thephage, the antigen would not be retained on the membrane.In this way, the loss of the tag sequence from the phage byproteolytic digestion could be monitored, and the efficacy ofcleavage of the substrate sequence could thereby be deter-mined. Using this dot blot assay, Smith et al. (9) were ableto compare quantitatively the cleavage rates of the selectedsequences. By monitoring the time-dependent loss of the tagsequence from the phage during proteolytic digestion, therelative concentration of the phage retaining the tag sequencefPg was determined at different time points. A set of slopescorresponding to first-order decay rates was obtained by plot-ting logfPg vs. time for each of the isolated clones. Thus, theseinvestigators concluded that the relative kcat=Km valuescould be determined by comparing the decay rates of theindividual clones.

292 Ohkubo

Sharkov et al. (33) monitored the hydrolysis of individualphage substrates by an ELISA. Isolated phage were incubatedwith protease and placed in microtiter plate wells, which werecoated with an antitag antibody. The captured phage werethen detected with an antiphage antibody. Sharkov et al.(33) characterized the reaction kinetics of the stromelysin pro-tease in this way. The phage can also be first immobilized inmicrotiter plate wells, then cleaved by the protease and theremaining tag sequence can be detected by an antitag anti-body (27). In general, the enzyme is present in excess as com-pared to the substrate during proteolytic digestion of substratephage (6,33), resulting in a single-turnover reaction, in whichthe equations for steady-state kinetics are not considered to bevalid. Sharkov et al. also showed that the proteolysis of sub-strate phage was a single-exponential process, and single-exponential rate constants appeared linear with respect tothe enzyme concentration throughout the range examined.Thus, these investigators concluded that the reaction betweenan enzyme and its substrate phage obeys the rules of pseudo-first-order kinetics, suggesting that time of incubation and theamount of enzyme are critical factors in designing substratephage selection experiments. They further suggested thatfew substrates would be found if the selection conditions weretoo stringent, i.e., if the incubation time was too short or theenzyme concentration too low. On the other hand, this techni-que would offer little discrimination among good substrates ifthe conditions were too relaxed (33). Thus, it is critical to opti-mize the conditions of substrate phage selection when usingthis approach to identify specific substrates.

III.D. Subtracted Substrate Phage Library

Tissue-type plasminogen activator (t-PA) and urokinase-typeplasminogen activator (u-PA) are members of the chymotryp-sin gene family that shares a high degree of structural similar-ity (34,35). A highly efficient substrate sequence for t-PA wasidentified by substrate phage display, and was found to becleaved up to 5300 times more efficiently by t-PA than wasthe primary sequence of the actual target site in plasminogen


itself (2,14). However, this newly identified sequence was alsocleaved efficiently by u-PA, and the selectivity of this substratewas only 4.7-fold higher for t-PA than for u-PA (2,14). To selectsubstrates specific for t-PA, Madison and colleagues (14)developed a novel protocol, which included a subtraction stepduring the phage selection, that removed substrates that werecleaved efficiently by u-PA. As a first step, a random substratephage library was subjected to high stringency selection witht-PA, to generate an intermediate library enriched in phagethat are efficient substrates of t-PA. This intermediate librarywas then digested at low stringency with u-PA to removephage that were moderate or good substrates for u-PA. Thephage that remained undigested with u-PA were defined asa subtracted substrate phage library, and the phage specificfor t-PA were recovered by digesting the subtracted substratephage library with t-PA. One of the highly selective substratesisolated from the subtracted substrate phage library showed47-fold higher catalytic efficiency for t-PA than that for u-PA(Q-R-G-R-K-A at the P4-P20 sites�). Comparison of the aminoacid sequences of substrates, derived from the subtracted sub-strate phage library and the standard substrate phage library,suggested that the P3 and P4 residues play a critical role indetermining the specificity between t-PA and u-PA for a givensubstrate. Whereas plasminogen activator inhibitor type I(PAI-I) is a primary physiological inhibitor of both enzymes,mutation of valine to glutamine at P4 and serine to arginineat P3 enhanced the specificity of PAI-I for t-PA by approxi-mately 600-fold (14).

IV. APPLICATION OF SUBSTRATE PHAGEDISPLAY TO CANCER RESEARCH

IV.A. Angiogenesis

Angiogenesis is the biological process by which new capillariesare formed from pre-existing blood vessels (36). This occurs

� The nomenclature for the substrate amino acids preference is Pn,Pn�1, . . . , P2, P1, P10, P20, . . . , Pm�10, Pm0. Amide bond hydrolysis occursbetween P1 and P10.

294 Ohkubo

under both physiological and pathological conditions. Forexample, the transition from an avascular to a vascularizedstate is considered to be a critical turning point in the develop-ment of tumors. Vascularization provides oxygen and essen-tial nutrients to the tumor and increases the proliferationrate of cancer cells. It has been established that a tumor masscannot exceed a size of �1 mm3 in an avascular state (37,38).Even though virtually every cancer exhibits a slightly differ-ent phenotype, the tumor endothelium is relatively uniformin all solid tumors. Thus, angiogenesis is regarded as a com-mon and key target for cancer chemopreventive agents.

Tumor angiogenesis depends mainly on the release ofspecific growth factors from neoplastic cells. These growthfactors bind to receptor tyrosine kinases (RTKs) expressed onthe endothelial cell surface. Binding of growth factors leadsto the phosphorylation and activation of RTK, which even-tually results in endothelial cell recruitment and prolifera-tion (36,39). Expression of vascular endothelial growthfactor (VEGF) is widely induced during angiogenesis, render-ing it a prime target for antivascular therapy (40). VEGF hasfive isoforms that exist as homodimers, and bind to the fms-like tyrosine kinase (flt-1) and fetal liver kinase (flk-1) recep-tors. The major known physiological function of VEGF is topromote angiogenesis in response to hypoxia. Several inhibi-tors of VEGF and VEGF receptors have now reached thestage of clinical trials (36,39). In addition, Tie-2 (tyrosinekinase with immunoglobulin and epidermal growth factorhomology domain) is an endothelium-specific RTK that bindsthe angiopoietin ligands, and plays an indispensable role invascular remodeling and maturation (41). Angiopoietinsseem to function in a complementary and coordinated fashionwith VEGF (42). It has been reported that increased Tie-2expression is associated with the growth of several types oftumors (43–45), and the growth of experimental tumors couldbe inhibited by blocking the Tie-2 pathway (46–48). Thesedata suggest that inhibitors of Tie-2 also may be useful asantiangiogenic cancer drugs.

Using substrate phage display, Deng et al. (32) haveidentified the substrate sequences of Tie-2 to develop


high-throughput screens for Tie-2 inhibitors. The phagelibraries consisted of sequences in the X-X-X-Y-X-X-X-X orX-X-X-X-Y-X-X-X-X motifs, where Y represents tyrosine andX represents a random mixture of all 20 amino acids. The pep-tide library was incubated with the catalytic domain of Tie-2,and phosphorylated phage particles were captured with anantiphosphotyrosine antibody and the peptide sequences wereanalyzed. Amongst four identified substrate sequences, R-L-V-A-Y-E-G-W-V exhibited the best catalytic efficiency with akcat=Km of 5.9�104 M�1 sec�1. This activity was sufficient todevelop a number of HTS assays, including dissociation-enhanced lanthanide fluoroimmunoassays (DELFIA), radioac-tive plate binding (RPB), and time-resolved fluorescent reso-nance energy transfer (TR-FRET). Automated DELFIA andTR-FRET assays were later developed, which have been suc-cessfully used to screen a combined set of >600,000 smallorganic molecules against Tie-2 (32).

IV.B. Prognostic Markers for Prostate Cancer

The human kallikrein family of serine proteases now includes15 family members that share significant homology (49,50).Human glandular kallikrein 3 (hK3), or prostate-specific anti-gen (PSA), is considered to be a prognostic marker for pros-tate cancer, and serum PSA levels are used for screeningand early detection of prostate cancer (51–53). In addition toPSA, human kallikrein 2 (hK2) has recently emerged as acomplementary marker for detecting prostate cancer, espe-cially at low PSA values (54–56). PSA and hK2 are the mosthighly homologous members of the human kallikrein family,with 78% and 80% identity at the amino acid and DNA levels,respectively (50). Whereas hK2 has a trypsin-like specificityfor arginine and lysine, PSA more closely resembles othertissue kallikreins, exhibiting a chymotrypsin-like preferencefor tyrosine and leucine (57,58). PSA hydrolyzes semenogelinI, semenogelin II, and fibronectin in vivo, and plays a role ofsemen liquefaction, a biological process which immediatelyfollows ejaculation (59–62). A number of other potential sub-strates for PSA have been identified, including TGF-b (63),

296 Ohkubo

parathyroid hormone-related protein (64) and insulin-likegrowth factor binding proteins (IGF-BPs) (65). Human kallik-rein 2 has also been shown to play a possible role in the earlystage of semen liquefaction and to have hydrolyzing activitytowards fibronectin and semenogelins (66). In addition, hK2can activate u-PA (67), inactivate PAI-I (68), and cleaveIGF-BPs (65). However, a role of these enzymes in cancerdevelopment has not yet been clearly defined. Characteriza-tion of the substrate specificity of PSA and hK2 may shedsome light on their roles during tumor progression and onadditional physiological functions of these enzymes. More-over, sensitive activity-based assays of these enzymes areuseful for monitoring prostate cancer diagnosis.

By using two independent approaches, substrate phagedisplay and iterative optimization of the synthetic peptidesderived from native substrate sequences of semenogelin, aconsensus substrate sequence for PSA was defined as S-S-(Y=F) -Y-S-G at the P4–P20 sites (3). These sequences werecleaved by PSA with catalytic efficiencies (kcat=Km) as highas 2200–3100 M�1 sec�1, as compared with values of2–46 M�1 sec�1 for peptides derived from the physiological tar-get sequences of semenogelin. The optimized consensus sub-strate sequence (S-S-Y-Y-S-G at the P4–P20 sites) does notfit into the known structural model for PSA, which has beendeposited in the Protein Data Bank (1PFA.PDB). Thus, anew three-dimensional model for PSA was reconstructedbased on the known structure of porcine tissue kallikrein(3). This new model indicates that a tyrosine residue at P1 ispreferred and suggests that the P2 residue of a substrate orinhibitor can be docked into a pocket formed by a large inser-tion loop.

The substrate specificity of hK2 was also characterizedusing a random pentapeptide phage library (22). After eightrounds of selection, genes encoding the random substratewere subcloned into an expression vector, to generate CFP-XXXXX-HHHHHH fusion proteins. These fusion proteinswere immobilized onto Ni-NTA beads and the time course ofsubstrate hydrolysis was followed by monitoring the fluores-cence released from the beads after incubation with hK2.


Thirty peptides were selected by phage screen with catalyticefficiencies (kcat=Km) for cleavage by hK2 ranging from1.7� 104 M�1 sec�1 for L-R-S-R-A at the P2–P30 sites to9.9� 101 M�1 sec�1 for E-R-V-S-P at the P2–P30 sites. Thedata show that hK2 cleaves quite selectively after arginineresidues, which is consistent with previous reports (69,70),and that hK2 specificity is further enhanced by a serine inthe P10 position. A SwissProt database search with selectedsequences identified three putative substrates: a disintegrin-like and metalloprotease domain with a thrombospondin typeI modules 8 (ADAM-TS8) precursor, a cadherin-related tumorsuppressor homolog, and a collagen a (IX) chain precursor. Itis plausible that the cleavage of these proteins by hK2 couldplay a role in cancer progression (71–74).

IV.C. Tumor Invasion and Metastasis

Matrix metalloproteinases (MMPs) are a family of zincendopeptidases capable of degrading extracellular matrix(ECM). These enzymes are essential for embryonic develop-ment, morphogenesis, reproduction, tissue resorption, andremodeling (75–78). Matrix metalloproteinases also partici-pate in various pathological processes, including rheumatoidarthritis, osteoarthritis, periodontitis, autoimmune blister-ing disorders of the skin, and tumor invasion and metasta-sis (75–78). To date, the MMP family includes 21 knownenzymes that are categorized by their domain structureand by their preferences for macromolecular substrates(Table 3) (75–78). The catalytic domain of MMPs is com-prised of a five-stranded b-sheet, three a-helices, and brid-ging loops, which forms a backbone structure that ishighly conserved among MMP family members (79–85). X-ray crystallographic analysis showed that the S10 subsitein MMPs, which is the most well-defined binding area,forms a hydrophobic pocket (79,82,83). The substrate speci-ficities of MMPs have been studied using collagen sequence-based synthetic peptides, which are natural substrates ofMMPs. This topic was reviewed by Nagase and Fields in1996 (86).

298 Ohkubo

Table 3 Human Matrix Metalloproteinase Family

Enzyme Known substrates

CollagenasesMMP-1 Collagenase-1 Collgen types, I, II, III, VII, VIII, X,

aggrecan, entactin, a1-PI, IL-1beta,TFPI, IGF-BP3, SAA, AFPs,proMMP-2

MMP-8 Collagenase-2 Collagen types I, II, III, aggrecan,substance-P, TFPI

MMP-13 Collagenase-3 Collagen types I, II, III, IV, IX, X,XIV, aggrecan, FN, gelatin, casein,tenascin, proMMP-9, osteonectin

GelatinasesMMP-2 Gelatinase A

(72 kDa)Collagens types I, IV, V, X, IX, XI,

proMMP-9, 13, Eph B1, aggrecan,VN, SAA, AFPs, APP, galectin-3tenascin, IL-1beta, decorin,osteonectin

MMP-9 Gelatinase B(92 kDa)

Collagens types I, III, IV, V, XI, XIV,XVII, entactin, a1-PI, galectin-3,substance-P, VN, aggrecan,IL-1beta, osteonectin, elastin,plasminogen, TFPI, gelatin, myelinbasic protein

StromelysinsMMP-3 Stromelysin-1 ProMMP-1, 8, 9, 13, collagen types I,

II, III, IV, IX, aggrecan, a1-PI, IGF-BP3, FN, laminin, casein, gelatin,tenascin, osteopontin, transferring,a2-macroglobulin, VN, osteonectin,fibrinogen, fibrin, IL-1beta,decorin, elastin, PAI-I, scu-PA,SAA, AFPs

MMP-10 Stromelysin-2 Collagen type IV, proMMP-1,-7, -8,-9,proteoglycan, gelatin, casein,aggrecan

MatrilysinsMMP-7 Matrilysin Aggrecan, entactin, aI-PI,

proMMP-2, decorin, TFPI, VN,carboxymethylated transferring,osteopontin, tenascin, casein,collagens types I, IV, XVIII,

(Continued)


Table 3 Human Matrix Metalloproteinase Family (Continued )

Enzyme Known substrates

E-cadherin, FN, fibulin,osteonectin, beta4-integrin,proTNF-a, plasminogen,fibrinogen, fibrin, FasL

MMP-26 Matrilysin-2 Collagen type IV, FN, fibrinogen,gelatin, a1-PI: proMMP9

Membrane-type MMPsMMP-14 MTI-MMP ProMMP2, proMMP-13, FN,

tenascin, nidogen, perlecan,collagen types I, II, III, VN, laminin,tTG, a2-microglobulin, aggrecan,fibrinogen, fibrin

MMP-15 MT2-MMP ProMMP2, tTG, FN, tenascin,nidogen, aggrecan, perlecan,laminin

MMP-16 MT3-MMP ProMMP2, collagen type III, gelatin,laminin-1, aggrecan, FN, VN,a1-PI, a2-macroglobulin, tTG

MMP-17 MT4-MMP Gelatin, pro-TNF-a, FN, fibrinMMP-24 MT5-MMP ProMMP-2, chondroitin sulfate

proteoglycan, FN, dematin sulfateproteoglycan

MMP-25 MT6-MMP ProMMP-2, collagen type IV, gelatin,FN, fibrin

OthersMMP-11 Stromelysin-3 a1-PI, serine protease inhibitor,

IGFBP-1, weak activity for ECMproteins

MMP-12 Metalloelastase Elastin, collagen types I, IV, V,osteonectin, FN, VN, gelatin,laminin, pro-TNF-a, TFPI, myelinbasic protein, a1-antitrypsin

MMP-19 RASI-1 Collagen type IV, laminin, nidogen,fibronectin, gelatin, COMP,aggrecan

MMP-20 Enamelysin Amelogenin, aggrecan, COMPMMP-23 CA-MMP UnknownMMP-28 Epilysin Casein

SAA: acute-phase serum amyloid A; AFPs: AA amyloid fibril proteins; scu-PA: single-chain urokinase-type plasminogen activator; APP: amyloid protein precursor; PAI-I:plasminogen activator inhibitor I; FN: fibronectin; VN: vitronectin; a1-PI: a1-protei-nase inhibitor; TEPI: tissue factor pathway inhibitor; tTG: tissue transglutaminase;ECM: extracellular matrix; COMP: cartilage oigomeric matrix protein.

300 Ohkubo

Substrate phage display has also been used to char-acterize the substrate specificities of MMPs, includingMMP-2, -3, -7, -9, -13, and -14 (MT1-MMP) (9,21,26–30).Smith et al. (9) compared the substrate specificity betweenstromelysin 1 (MMP-3) and matrilysin (MMP-7) by using arandom hexamer phage library. While both enzymes favoredproline at the P3 position, which is frequently found in sub-strates for other MMPs (86), differences were observed inthe preferences of stromelysin 1 and matrilysin at the P2and P10 positions of the substrates. The substitution of leucinefor methionine at the P10 resulted in a modest increase instromelysin 1 activity, but decreased matrilysin activity by8-fold. The substitution of leucine for phenylalanine at P2resulted in a 3-fold decrease in the catalytic efficiency of stro-melysin 1, but a nearly 3-fold increase in matrilysin activity.

Highly selective and efficient substrates of humancollagenase 3 (MMP-13) were identified by the substratephage technique (28). A consensus substrate sequence for col-lagenase 3, at the P3–P30 sites (P-L-G-M-R-G) was deducedbased on the preferred residue in each subsite position from35 selected phage clones. A synthetic peptide correspondingto the consensus sequence (G-P-L-G-M-R-G-L) exhibited ahigher kcat=Km value (4.2� 106 M�1 sec�1) for hydrolysis bycollagenase 3, and was a more efficient substrate than thepreviously reported substrates, such as McaPChaGNva-HADpa-NH2 and McaPLGLDpaAR-NH2 (87). The catalyticefficiency of collagenase 3 for hydrolysis of the consensussequence was 1344-, 11-, and 820-fold higher than those forstromelysin 1 (MMP-3), gelatinase B (MMP-9), and collage-nase 1 (MMP-1), respectively. Substitution at the P3 residuerevealed that collagenase 3 also favors proline as a P3 resi-due. A search of the SwissProt and translated EMBL proteindatabases with selected phage sequences identified potentialcollagenase 3 cleavage sites in type IV collagen, biglycan,and the latency-associated peptide of TGF-b3. This match isconsistent with the role of MMPs in proteolytic activation ofTGF-b3 (88).

The MMPs have been implicated in the process of tumorgrowth, invasion, and metastasis (89–92). Among the MMPs,


the gelatinases (MMP-2 and MMP-9), which can both degradecomponents of basement membranes, have been mostconsistently detected in malignant tissues and associatedwith tumor aggressiveness, metastatic potential, and a poorprognosis (78,89,93–95). Most MMPs are secreted as latentprecursors (zymogens) and are subsequently activated byproteolysis. Membrane type-1 matrix metalloproteinase(MT1-MMP) has been cloned as an activator of MMP-2 (96).It has been reported that MT1-MMP is also overexpressedin several types of tumors (97–103), suggesting that theactivation of MMP-2 by MT1-MMP may play a critical rolein cancer cell invasion and metastasis. On the other hand,MT1-MMP itself can digest many types of ECM proteins, suchas interstitial collagens, gelatin, and proteoglycans (104–106).These results suggest that MT1-MMP plays a dual role in thedigestion of ECM, by both directly cleaving the substrate andby activating MMP-2. Since the relationships between MMPsand cancer invasion and metastasis have been delineated,several MMP inhibitors (MMPIs) have been developed andare expected to represent a new approach to cancer treat-ment, in addition to the traditional cytotoxic drugs (78,95).However, many of these MMPIs are broad-spectrum inhibi-tors, and exhibit undesirable side effects (78,95). The princi-pal side effect of these drugs is musculoskeletal pain,suggesting that for cancer treatment, it will be important todevelop selective gelatinase inhibitors with limited activityagainst collagenases involved in the maintenance of normaljoint function.

The substrate sequence specificity of MT1-MMP wasanalyzed using a random hexamer phage library (21,29). Byaligning the selected clones, the consensus substrate sequencefor MT1-MMP was defined as P-X-(G=P) -(L=I) at the P3–P10

sites. The deduced consensus substrate sequence resemblesthe canonical collagen-like P-X-X-L motif, which has been pre-viously established for MMPs (86). In fact, the P-X-G-L=Isequence was reported to be present in human type I collagen(a1)(P-Q-G-I), type I collagen (a2)(P-Q-G-L), type II collagen(P-Q-G-L), type III collagen (P-L-G-I), and a2-macroglobulin(P-E-G-L), and these proteins were susceptible to digestion

302 Ohkubo

by MT1-MMP in vitro (104). The synthetic peptide preparedbased on the consensus sequence (G-P-L-G-L-R-S-W from P4to P40) was cleaved efficiently by MT1-MMP; however, thispeptide was also efficiently hydrolyzed by MMP-2 and MMP-9.

Highly selective substrate sequences for MMP-2, MMP-9, and MT1-MMP were identified in a series of substratephage studies by Smith and coworkers (26,27,30). In eachscreen, the canonical MMP substrate sequence, P-X-X-XHy

(where XHy is a hydrophobic residue) at the P3–P10 sitesemerged as a consensus sequence. However, other specificsubstrate sequences for individual MMPs were also identifiedfrom the selected clones. Substrate phage display was used toidentify substrate sequences of MMP-2 and these were cate-gorized into four distinct groups, based on their sequencesimilarities (26). Whereas the substrates containing theP-X-X-XHy canonical motif lacked selectivity, the other threegroups contained a unique consensus sequence and showedhigher selectivity for MMP-2 over the other MMPs tested.These substrates contained consensus motifs of (L=I)-X-X-XHy, XHy-S-X-L, and H-X-X-XHy at the P3–P10 sites.Among these sequences, the (L=I) -X-X-XHy peptides exhibitedthe most selectivity, and one of the highly selective substrates(S-G-R-S-L-S-R-L-T-A at the P7–P30 sites) was cleaved200-fold more efficiently by MMP-2 than by its closely relatedhomolog, MMP-9. Though this motif does not contain a pro-line at the P3 position, which frequently occurs in MMP sub-strates, substitution at the P3 position of (L=I) -X-X-XHy

peptide revealed that the absence of proline at this positionis not the sole determinant for MMP-2 selectivity. Rather,the P2 position plays a major role in determining specificitybetween MMP-2 and MMP-9. Using substrate phage display,the consensus substrate sequence for MMP-9 was character-ized as P-R-(S=T) -XHy-(S=T) at the P3–P20 sites, and the pep-tides containing arginine at the P2 position were cleaved mostefficiently by MMP-9 (27). Since arginine was rarely presentat P2 in the substrates selective for MMP-2, arginine wassubstituted at this position within the MMP-2 selective pep-tides, and catalytic activities were determined. Interestingly,the substitution of arginine for serine at the P2 position


dramatically increased hydrolysis by MMP-9, and signifi-cantly decreased hydrolysis by MMP-2. These observationssuggest that the interaction between the P2 position of thesubstrate and the S2 position within the catalytic cleft ofMMP plays a key role in distinguishing substrate recognitionby MMP-2 and MMP-9. Indeed, analysis of the structure of theS2 subsite within MMP-2 and MMP-9 identified a potentialstructural basis for this distinction in substrate recognitionat the P2 position (26).

The MT1-MMP selective phage were isolated from sub-strate phage clones by comparing the catalytic activity ofMT1-MMP, MMP-2, and MMP-9 for individual phage (30).No consensus sequence could be defined from the MT1-MMPselective clones; however, the residues that include long sidechains, particularly arginine, were favored at the P4 position.In fact, the substitution of alanine for arginine at P4 resultedin substrates that were poorly cleaved by MT1-MMP. Like thecanonical substrate sequences for MMPs, a hydrophobic resi-due was preferred at the P10 position in MT1-MMP selectivesubstrates. Interestingly, proline, which is the favored residueat the P3 position among the various substrates of MMPs, isabsent from these sequences. The peptide derived from phageclone (S-G-R-S-E-N-I-R-T-A at the P6–P40 sites) was hydro-lyzed 83-fold more efficiently by MT1-MMP than by MMP-9.Hence, the substitution of serine for proline at the P3 positionof the above sequence converted a substrate selective for MT1-MMP to a substrate that is recognized equally well by bothenzymes. The idea that MT1-MMP recognizes substrates intwo distinct modes arose from these observations (30). Onemode makes use of the P3 and P10 positions as dominant con-tact points that bind to nonselective substrates. In the othermode, the P4 and P10 subsites appear to form contacts thatare critical for recognizing specific substrates. Indeed, athree-dimensional modeling of the selective and nonselectivesubstrates bound to the catalytic pocket of MT1-MMP couldexplain two separate binding modes. These new insights canbe used to help design highly selective inhibitors of individualMMPs, and may also provide additional clues for understand-ing the physiological roles of MMPs.

304 Ohkubo

V. CONCLUSIONS

As we have discussed in this chapter, substrate phage displayhas enormous potential for delineating the substrate specifici-ties of proteases and protein kinases. This information mayfacilitate the development of potent and selective inhibitorsfor improved therapeutics. Knowledge of the sequence specifi-cities of enzymes, especially of closely related enzymes, mightserve as a template for understanding structure–activity rela-tionships and for the rational design of new drugs targetingthese enzymes. The selective substrates can be converted intoprobes to be used in HTS assays. In addition, the results ofsubstrate phage display can be used for other applications,such as activity-based measurement of enzymes and predic-tion of the physiological and pathophysiological substratesfor enzymes. Indeed, several novel potential protease sub-strates have been successfully identified using substratephage display (Table 4).

Though substrate phage display has been applied toscreen for substrates of proteases and kinases, one wouldpredict that this technique could also be applied to identifysubstrates for other enzymes. Enzymatic modifications of pro-teins that can be carried out in vitro, including acetylation,ubiquitination, and glycosylation, should be good candidatesfor applying substrate phage display. It is also conceivablethat, with modification, a similar strategy could be appliedto screen for substrates of protein phosphatases. Dente etal.(31) have shown that, by extending the kinase reactiontime, the sequence specificity is weakened and practicallyany tyrosine-containing sequence can become phosphory-lated. Thus, it is possible to make a modified phage peptidelibrary, where all the tyrosine residues are converted to phos-photyrosine in vitro. Such a modified phage library could thenbe used to identify specific substrates for protein tyrosinephosphatases.

As of August 2002, 498 peptidase sequences selectedfrom the primary databases were deposited in the MEROPSpeptidase database (107). It is expected that some of these willbecome potential new drug targets (108). It is now possible to


Table 4 Novel Potential Substrate Proteins Identified by Substrate Phage Display

Protease

Substratesequenceidentified Potential substrate and target sequence Reference

Rat granzyme B IEXD#XG Poly (ADP-ribose) polymerase (PARP) (18)(VDPD#SG, LEID#YG)

Pro-caspase 3 (IETD#SG)Pro-caspase 7 (IQAD#SG)

Human membrane- (R=K)XSR#A Protease-activated receptor (PAR) 2 (SKGR#S) (20)type serineprotease 1

X(R=K)SR#A Single-chain urokinase-type plasminogenactivator(sc-uPA) (PREK#)

Human collagenase 3 GPLG#MRGL Biglycan (PKG#VFS) (28)(MMP-13) TGF-b3 (PKG#ITS)

Human gelatinase B PR(S=T)# Kallikrein 14 (PRT#IT) (27)(MMP-9) XHy(S=T) Ladinin 1 (PRT#IS)

Endoglin (PRT#VT)Endothelin receptor (PRT#IS)Laminin a3 chain (PRS#LT)Phosphate regulating neutral endopeptidase

(PRS#LS)ADAM 2 (PRT#IS)Desmoglein 3 (PRS#LT)Integlin b5(PRS#IT)

306

Ohku

bo

Rat mast cell protease 4 LVWF#RG Protein C precursor (VVFF#RG) (23)(rMCP-4) Procollagen C-proteinase enhancer protein

(LLWY#SG)Coagulation factor V (VMYF#NG)TGF-b receptor type III (VVYY#NS)Cystein-rich secretory protein-3 (VVWY#SS)Plasminogen activator inhibitor-1 (ALYF#NG)Low affinity Igg Fc Region receptor III (LVWF#HA)

Human kallikrein 2 LR#SRA ADAM-TS 8 precursor (RGR#SE) (22)(hK2) Cadherin-related tumor suppressor homologue

precursor (GVFR#S)Collagen a (IX)chain precursor (PGR#AP)

Human gelatinase A(MMP-2)

XHySX#La Eph B1 tyrosine kinase receptor (YKSE#LRE) (26)

aXHy is a hydrophobic residue,#site of digestion.

Substrate

Phage

Disp

lay307

quickly determine the activities and substrate specificities ofthese proteases by substrate phage display. This techniquewill undoubtedly contribute to molecular medicine and thedevelopment of novel therapeutic strategies in the future.

REFERENCES

1. Zhou S, Cantley LC. Recognition and specificity in proteintyrosine kinase-mediated signalling. Trends Biochem Sci1995; 20:470–475.

2. Ding L, Coombs GS, Strandberg L, Navre M, Corey DR,Madison EL. Origins of the specificity of tissue-type plasmi-nogen activator. Proc Natl Acad Sci USA 1995; 92:7627–7631.

3. Coombs GS, Bergstrom RC, Pellequer JL, Baker SI, Navre M,Smith MM, Tainer JA, Madison EL, Corey DR. Substratespecificity of prostate-specific antigen (PSA). Chem Biol1998; 5:475–488.

4. Hervio LS, Coombs GS, Bergstrom RC, Trivedi K, Corey DR,Madison EL. Negative selectivity and the evolution ofprotease cascades: the specificity of plasmin for peptide andprotein substrates. Chem Biol 2000; 7:443–453.

5. Matthews DJ, Wells JA. Substrate phage: selection of protease sub-strates by monovalent phage display. Science 1993; 260:1113–1117.

6. Kay BK, Winter J, McCafferty J. Phage Display of Peptidesand Proteins. London: Academic Press, 1996.

7. Gram H. Phage display in proteolysis and signal transduc-tion. Comb Chem High Throughput Screen 1999; 2:19–28.

8. Smith GP, Scott JK. Libraries of peptides and proteins displayedon filamentous phage. Methods Enzymol 1993; 217:228–257.

9. Smith MM, Shi L, Navre M. Rapid identification of highlyactive and selective substrates for stromelysin and matrilysinusing bacteriophage peptide display libraries. J Biol Chem1995; 270:6440–6449.

10. Schmitz R, Baumann G, Gram H. Catalytic specificity ofphosphotyrosine kinases Blk, Lyn, c-Src and Syk as assessedby phage display. J Mol Biol 1996; 260:664–677.

308 Ohkubo

11. Westendorf JM, Rao PN, Gerace L. Cloning of cDNAs forM-phase phosphoproteins recognized by the MPM2 monoclo-nal antibody and determination of the phosphorylatedepitope. Proc Natl Acad Sci USA 1994; 91:714–718.

12. Ballinger MD, Tom J, Wells JA. Designing subtilisin BPN0 tocleave substrates containing dibasic residues. Biochemistry1995; 34:13312–13319.

13. Matthews DJ, Goodman LJ, Gorman CM, Wells JA. A surveyof furin substrate specificity using substrate phage display.Protein Sci 1994; 3:1197–1205.

14. Ke SH, Coombs GS, Tachias K, Navre M, Corey DR, MadisonEL. Distinguishing the specificities of closely related pro-teases. Role of P3 in substrate and inhibitor discriminationbetween tissue-type plasminogen activator and urokinase. JBiol Chem 1997; 272:16603–16609.

15. Ke SH, Coombs GS, Tachias K, Corey DR, Madison EL. Opti-mal subsite occupancy and design of a selective inhibitor ofurokinase. J Biol Chem 1997; 272:20456–20462.

16. O’Boyle DR II, Pokornowski KA, McCann PJ III, WeinheimerSP. Identification of a novel peptide substrate of HSV-1 pro-tease using substrate phage display. Virology 1997; 236:338–347.

17. Huang C, Wong GW, Ghildyal N, Gurish MF, Sali A,Matsumoto R, Qiu WT, Stevens RL. The tryptase, mousemast cell protease 7, exhibits anticoagulant activity in vivoand in vitro due to its ability to degrade fibrinogen in the pre-sence of the diverse array of protease inhibitors in plasma. JBiol Chem 1997; 272:31885–31893.

18. Harris JL, Peterson EP, Hudig D, Thornberry NA, Craik CS.Definition and redesign of the extended substrate specificityof granzyme B. J Biol Chem 1998; 273:27364–27373.

19. Dall’Acqua W, Halin C, Rodrigues ML, Carter P. Elastasesubstrate specificity tailored through substrate-assisted cata-lysis and phage display. Protein Eng 1999; 12:981–987.

20. Takeuchi T, Harris JL, Huang W, Yan KW, Coughlin SR,Craik CS. Cellular localization of membrane-type serine pro-tease 1 and identification of protease-activated receptor-2 and


single-chain urokinase-type plasminogen activator assubstrates. J Biol Chem 2000; 275:26333–26342.

21. Ohkubo S, Miyadera K, Sugimoto Y, Matsuo K, Wierzba K,Yamada Y. Substrate phage as a tool to identify novel sub-strate sequences of proteases. Comb Chem High ThroughputScreen 2001; 4:573–583.

22. Cloutier SM, Chagas JR, Mach JP, Gygi CM, Leisinger HJ,Deperthes D. Substrate specificity of human kallikrein 2(hK2) as determined by phage display technology. Eur JBiochem 2002; 269:2747–2754.

23. Karlson U, Pejler G, Froman G, Hellman L. Rat mast cell pro-tease 4 is a beta-chymase with unusually stringent substraterecognition profile. J Biol Chem 2002; 277:18579–18585.

24. Sharkov NA, Cai D. Discovery of substrate for type I signalpeptidase SpsB from Staphylococcus aureus. J Biol Chem2002; 277:5796–5803.

25. Beck ZQ, Hervio L, Dawson PE, Elder JH, Madison EL. Iden-tification of efficiently cleaved substrates for HIV-1 proteaseusing a phage display library and use in inhibitor develop-ment. Virology 2000; 274:391–401.

26. Chen EI, Kridel SJ, Howard EW, Li W, Godzik A, Smith JW.A unique substrate recognition profile for matrix metallopro-teinase-2. J Biol Chem 2002; 277:4485–4491.

27. Kridel SJ, Chen E, Kotra LP, Howard EW, Mobashery S,Smith JW. Substrate hydrolysis by matrix metalloprotei-nase-9. J Biol Chem 2001; 276:20572–20578.

28. Deng SJ, Bickett DM, Mitchell JL, Lambert MH, BlackburnRK, Carter HL III, Neugebauer J, Pahel G, Weiner MP, MossML. Substrate specificity of human collagenase 3 assessedusing a phage-displayed peptide library. J Biol Chem 2000;275:31422–31427.

29. Ohkubo S, Miyadera K, Sugimoto Y, Matsuo K, Wierzba K,Yamada Y. Identification of substrate sequences for mem-brane type-1 matrix metalloproteinase using bacteriophagepeptide display library. Biochem Biophys Res Commun1999; 266:308–313.

310 Ohkubo

30. Kridel SJ, Sawai H, Ratnikov BI, Chen EI, Li W, Godzik A,Strongin AY, Smith JW. A unique substrate binding modediscriminates membrane type-1 matrix metalloproteinasefrom other matrix metalloproteinases. J Biol Chem 2002;277:23788–23793.

31. Dente L, Vetriani C, Zucconi A, Pelicci G, Lanfrancone L,Pelicci PG, Cesareni G. Modified phage peptide libraries asa tool to study specificity of phosphorylation and recognitionof tyrosine containing peptides. J Mol Biol 1997; 269:694–703.

32. Deng SJ, Liu W, Simmons CA, Moore JT, Tian G. Identifyingsubstrates for endothelium-specific Tie-2 receptor tyrosinekinase from phage-displayed peptide libraries for highthroughput screening. Comb Chem High Throughput Screen2001; 4:525–533.

33. Sharkov NA, Davis RM, Reidhaar-Olson JF, Navre M, Cai D.Reaction kinetics of protease with substrate phage. Kineticmodel developed using stromelysin. J Biol Chem 2001; 276:10788–10793.

34. Spraggon G, Phillips C, Nowak UK, Ponting CP, Saunders D,Dobson CM, Stuart DI, Jones EY. The crystal structure of thecatalytic domain of human urokinase-type plasminogen acti-vator. Structure 1995; 3:681–691.

35. Lamba D, Bauer M, Huber R, Fischer S, Rudolph R, KohnertU, Bode W. The 2.3 A crystal structure of the catalytic domainof recombinant two-chain human tissue-type plasminogenactivator. J Mol Biol 1996; 258:117–135.

36. Ribatti D, Vacca A, Nico B, De Falco G, Giuseppe Montaldo P,Ponzoni M. Angiogenesis and anti-angiogenesis in neuroblas-toma. Eur J Cancer 2002; 38:750–757.

37. Woodhouse EC, Chuaqui RF, Liotta LA. General mechanismsof metastasis. Cancer 1997; 80:1529–1537.

38. Ferrara N, Alitalo K. Clinical applications of angiogenicgrowth factors and their inhibitors. Nat Med 1999; 5:1359–1364.

39. Dy GK, Adjei AA. Novel targets for lung cancer therapy: partII. J Clin Oncol 2002; 20:3016–3028.


40. Toi M, Matsumoto T, Bando H. Vascular endothelial growthfactor: its prognostic, predictive, and therapeutic implica-tions. Lancet Oncol 2001; 2:667–673.

41. Lauren J, Gunji Y, Alitalo K. Is angiopoietin-2 necessary forthe initiation of tumor angiogenesis? Am J Pathol 1998; 153:1333–1339.

42. Holash J, Wiegand SJ, Yancopoulos GD. New model of tumorangiogenesis: dynamic balance between vessel regression andgrowth mediated by angiopoietins and VEGF. Oncogene1999; 18:5356–5362.

43. Stratmann A, Risau W, Plate KH. Cell type-specific expres-sion of angiopoietin-1 and angiopoietin-2 suggests a role inglioblastoma angiogenesis. Am J Pathol 1998; 153:1459–1466.

44. Takahama M, Tsutsumi M, Tsujiuchi T, Nezu K, Kushibe K,Taniguchi S, Kotake Y, Konishi Y. Enhanced expression ofTie2, its ligand angiopoietin-1, vascular endothelial growthfactor, and CD31 in human non-small cell lung carcinomas.Clin Cancer Res 1999; 5:2506–2510.

45. Etoh T, Inoue H, Tanaka S, Barnard GF, Kitano S, Mori M.Angiopoietin-2 is related to tumor angiogenesis in gastric car-cinoma: possible in vivo regulation via induction of proteases.Cancer Res 2001; 61:2145–2153.

46. Lin P, Polverini P, Dewhirst M, Shan S, Rao PS, Peters K.Inhibition of tumor angiogenesis using a soluble receptorestablishes a role for Tie2 in pathologic vascular growth. JClin Invest 1997; 100:2072–2078.

47. Siemeister G, Schirner M, Weindel K, Reusch P, Menrad A,Marme D, Martiny-Baron G. Two independent mechanismsessential for tumor angiogenesis: inhibition of human mela-noma xenograft growth by interfering with either the vascu-lar endothelial growth factor receptor pathway or the Tie-2pathway. Cancer Res 1999; 59:3185–3191.

48. Lin P, Buxton JA, Acheson A, Radziejewski C, MaisonpierrePC, Yancopoulos GD, Channon KM, Hale LP, DewhirstMW, George SE, Peters KG. Antiangiogenic gene therapytargeting the endothelium-specific receptor tyrosine kinaseTie2. Proc Natl Acad Sci USA 1998; 95:8829–8834.

312 Ohkubo

49. Diamandis EP, Yousef GM, Clements J, Ashworth LK,Yoshida S, Egelrud T, Nelson PS, Shiosaka S, Little S, LiljaH, Stenman UH, Rittenhouse HG, Wain H. New nomencla-ture for the human tissue kallikrein gene family. Clin Chem2000; 46:1855–1858.

50. Yousef GM, Diamandis EP. The new human tissue kallikreingene family: structure, function, and association to disease.Endocr Rev 2001; 22:184–204.

51. Catalona WJ, Smith DS, Ratliff TL, Dodds KM, Coplen DE,Yuan JJ, Petros JA, Andriole GL. Measurement of prostate-specific antigen in serum as a screening test for prostatecancer. N Engl J Med 1991; 324:1156–1161.

52. Oesterling JE. Prostate specific antigen: a critical assessmentof the most useful tumor marker for adenocarcinoma of theprostate. J Urol 1991; 145:907–923.

53. Labrie F, Dupont A, Suburu R, Cusan L, Tremblay M, GomezJL, Emond J. Serum prostate specific antigen as pre-screeningtest for prostate cancer. J Urol 1999; 147:846–851; discussion851–852.

54. Kwiatkowski MK, Recker F, Piironen T, Pettersson K, Otto T,Wernli M, Tscholl R. In prostatism patients the ratio ofhuman glandular kallikrein to free PSA improves the discri-mination between prostate cancer and benign hyperplasiawithin the diagnostic ‘‘gray zone’’ of total PSA 4 to 10 ng=mL.Urology 1998; 52:360–365.

55. Partin AW, Catalona WJ, Finlay JA, Darte C, Tindall DJ,Young CY, Klee GG, Chan DW, Rittenhouse HG, Wolfert RL,Woodrum DL. Use of human glandular kallikrein 2 for thedetection of prostate cancer: preliminary analysis. Urology1999; 54:839–845.

56. Nam RK, Diamandis EP, Toi A, Trachtenberg J, Magklara A,Scorilas A, Papnastasiou PA, Jewett MA, Narod SA. Serumhuman glandular kallikrein-2 protease levels predict thepresence of prostate cancer among men with elevated pros-tate-specific antigen. J Clin Oncol 2000; 18:1036–1042.

57. Watt KW, Lee PJ, M’Timkulu T, ChanWP, Loor R. Humanprostate-specific antigen: structural and functional similarity


with serine proteases. Proc Natl Acad Sci USA 1986;83:3166–3170.

58. Akiyama K, Nakamura T, Iwanaga S, Hara M. Thechymotrypsin-like activity of human prostate-specific anti-gen, gamma-seminoprotein. FEBS Lett 1987; 225:168–172.

59. Lilja H. A kallikrein-like serine protease in prostatic fluidcleaves the predominant seminal vesicle protein. J ClinInvest 1985; 76:1899–1903.

60. Lilja H, Oldbring J, Rannevik G, Laurell CB. Seminal vesicle-secreted proteins and their reactions during gelation andliquefaction of human semen. J Clin Invest 1987; 80:281–285.

61. Lilja H. Structure and function of prostatic- and seminalvesicle-secreted proteins involved in the gelation and lique-faction of human semen. Scand J Clin Lab Invest Suppl1988; 191:13–20.

62. Robert M, Gagnon C. Semenogelin I: a coagulum forming,multifunctional seminal vesicle protein. Cell Mol Life Sci1999; 55:944–960.

63. Killian CS, Corral DA, Kawinski E, Constantine RI.Mitogenic response of osteoblast cells to prostate-specificantigen suggests an activation of latent TGF-beta and a pro-teolytic modulation of cell adhesion receptors. Biochem Bio-phys Res Commun 1993; 192:940–947.

64. Iwamura M, Hellman J, Cockett AT, Lilja H, Gershagen S.Alteration of the hormonal bioactivity of parathyroidhormone-related protein (PTHrP) as a result of limitedproteolysis by prostate-specific antigen. Urology 1996;48:317–325.

65. Rehault S, Monget P, Mazerbourg S, Tremblay R, Gutman N,Gauthier F, Moreau T. Insulin-like growth factor binding pro-teins (IGFBPs) as potential physiological substrates forhuman kallikreins hK2 and hK3. Eur J Biochem 2001; 268:2960–2968.

66. Deperthes D, Frenette G, Brillard-Bourdet M, Bourgeois L,Gauthier F, Tremblay RR, Dube JY. Potential involvementof kallikrein hK2 in the hydrolysis of the human seminal vesi-cle proteins after ejaculation. J Androl 1996; 17:659–665.

314 Ohkubo

67. Frenette G, Tremblay RR, Lazure C, Dube JY. Prostatickallikrein hK2, but not prostate-specific antigen (hK3), acti-vates single-chain urokinase-type plasminogen activator.Int J Cancer 1997; 71:897–899.

68. Mikolajczyk SD, Millar LS, Kumar A, Saedi MS.Prostatic human kallikrein 2 inactivates and complexeswith plasminogen activator inhibitor-1. Int J Cancer 1999;81:438–442.

69. Bourgeois L, Brillard-Bourdet M, Deperthes D, Juliano MA,Juliano L, Tremblay RR, Dube JY, Gauthier F. Serpin-derived peptide substrates for investigating the substratespecificity of human tissue kallikreins hK1 and hK2. J BiolChem 1997; 272:29590–29595.

70. Mikolajczyk SD, Millar LS, Kumar A, Saedi MS. Humanglandular kallikrein, hK2, shows arginine-restricted specifi-city and forms complexes with plasma protease inhibitors.Prostate 1998; 34:44–50.

71. Dunne J, Hanby AM, Poulsom R, Jones TA, Sheer D, ChinWG, Da SM, Zhao Q, Beverley PC, Owen MJ. Molecular clon-ing and tissue expression of FAT, the human homologue ofthe Drosophila fat gene that is located on chromosome4q34–q35 and encodes a putative adhesion molecule.Genomics 1995; 30:207–223.

72. Georgiadis KE, Hirohata S, Seldin MF, Apte SS. ADAM-TS8,a novel metalloprotease of the ADAM-TS family located onmouse chromosome 9 and human chromosome 11. Genomics1999; 62:312–315.

73. Wang SS, Virmani A, Gazdar AF, Minna JD, Evans GA.Refined mapping of two regions of loss of heterozygosity onchromosome band 11q23 in lung cancer. Genes ChromosomesCancer 1999; 25:154–159.

74. Muragaki Y, Kimura T, Ninomiya Y, Olsen BR. The completeprimary structure of two distinct forms of human alpha 1 (IX)collagen chains. Eur J Biochem 1990; 192:703–708.

75. Johnson LL, Dyer R, Hupe DJ. Matrix metalloproteinases.Curr Opin Chem Biol 1998; 2:466–471.


76. Shapiro SD. Matrix metalloproteinase degradation of extra-cellular matrix: biological consequences. Curr Opin Cell Biol1998; 10:602–608.

77. Nagase H, Woessner JF Jr. Matrix metalloproteinases. J BiolChem 1999; 274:21491–21494.

78. Vihinen P, Kahari VM. Matrix metalloproteinases in cancer:prognostic markers and therapeutic targets. Int J Cancer2002; 99:157–166.

79. Bode W, Reinemer P, Huber R, Kleine T, Schnierer S,Tschesche H. The X-ray crystal structure of the catalyticdomain of human neutrophil collagenase inhibited by a sub-strate analogue reveals the essentials for catalysis and speci-ficity. EMBO J 1994; 13:1263–1269.

80. Lovejoy B, Cleasby A, Hassell AM, Longley K, Luther MA,Weigl D, McGeehan G, McElroy AB, Drewry D, LambertMH, et al. Structure of the catalytic domain of fibroblastcollagenase complexed with an inhibitor. Science 1994; 263:375–377.

81. Stocker W, Grams F, Baumann U, Reinemer P, Gomis-RuthFX, McKay DB, Bode W. The metzincins—topological andsequential relations between the astacins, adamalysins, ser-ralysins, and matrixins (collagenases) define a superfamilyof zinc-peptidases. Protein Sci 1995; 4:823–840.

82. Grams F, Reinemer P, Powers JC, Kleine T, Pieper M,Tschesche H, Huber R, Bode W. X-ray structures of humanneutrophil collagenase complexed with peptide hydroxamateand peptide thiol inhibitors. Implications for substrate bind-ing and rational drug design. Eur J Biochem 1995; 228:830–841.

83. Dhanaraj V, Ye QZ, Johnson LL, Hupe DJ, Ortwine DF,Dunbar JB Jr, Rubin JR, Pavlovsky A, Humblet C, BlundellTL. X-ray structure of a hydroxamate inhibitor complex ofstromelysin catalytic domain and its comparison with mem-bers of the zinc metalloproteinase superfamily. Structure1996; 4:375–386.

84. Gomis-Ruth FX, Maskos K, Betz M, Bergner A, Huber R,Suzuki K, Yoshida N, Nagase H, Brew K, Bourenkov GP,Bartunik H, Bode W. Mechanism of inhibition of the human

316 Ohkubo

matrix metalloproteinase stromelysin-1 by TIMP-1. Nature1997; 389:77–81.

85. Morgunova E, Tuuttila A, Bergmann U, Isupov M, LindqvistY, Schneider G, Tryggvason K. Structure of human pro-matrix metalloproteinase-2: activation mechanism revealed.Science 1999; 284:1667–1670.

86. Nagase H, Fields GB. Human matrix metalloproteinasespecificity studies using collagen sequence-based syntheticpeptides. Biopolymers 1996; 40:399–416.

87. Knauper V, Lopez-Otin C, Smith B, Knight G, Murphy G.Biochemical characterization of human collagenase-3. J BiolChem 1996; 271:1544–1550.

88. Yu Q, Stamenkovic I. Cell surface-localized matrix metallo-proteinase-9 proteolytically activates TGF-beta and promotestumor invasion and angiogenesis. Genes Dev 2000; 14:163–176.

89. Stetler-Stevenson WG, Aznavoorian S, Liotta LA. Tumor cellinteractions with the extracellular matrix during invasionand metastasis. Annu Rev Cell Biol 1993; 9:541–573.

90. Chambers AF, Matrisian LM. Changing views of the role ofmatrix metalloproteinases in metastasis. J Natl Cancer Inst1997; 89:1260–1270.

91. Kahari VM, Saarialho-Kere U. Matrix metalloproteinasesand their inhibitors in tumour growth and invasion. AnnMed 1999; 31:34–45.

92. Kleiner DE, Stetler-Stevenson WG. Matrix metalloprotei-nases and metastasis. Cancer Chemother Pharmacol 1999;43(suppl):S42–S51.

93. Noel A, Emonard H, Polette M, Birembaut P, Foidart JM.Role of matrix, fibroblasts and type IV collagenases in tumorprogression and invasion. Pathol Res Pract 1994; 190:934–941.

94. Ray JM, Stetler-Stevenson WG. The role of matrix metallo-proteases and their inhibitors in tumour invasion, metastasisand angiogenesis. Eur Respir J 1994; 7:2062–2072.


95. Hidalgo M, Eckhardt SG. Development of matrix metallopro-teinase inhibitors in cancer therapy. J Natl Cancer Inst 2001;93:178–193.

96. Sato H, Takino T, Okada Y, Cao J, Shinagawa A, YamamotoE, Seiki M. A matrix metalloproteinase expressed on thesurface of invasive tumour cells. Nature 1994; 370:61–65.

97. Okada A, Bellocq JP, Rouyer N, Chenard MP, Rio MC, Cham-bon P, Basset P. Membrane-type matrix metalloproteinase(MT-MMP) gene is expressed in stromal cells of human colon,breast, and head and neck carcinomas. Proc Natl Acad SciUSA 1995; 92:2730–2734.

98. Nomura H, Sato H, Seiki M, Mai M, Okada Y. Expression ofmembrane-type matrix metalloproteinase in human gastriccarcinomas. Cancer Res 1995; 55:3263–3266.

99. Tokuraku M, Sato H, Murakami S, Okada Y, Watanabe Y,Seiki M. Activation of the precursor of gelatinase A=72 kDatype IV collagenase=MMP-2 in lung carcinomas correlateswith the expression of membrane-type matrix metalloprotei-nase (MT-MMP) and with lymph node metastasis. Int JCancer 1995; 64:355–359.

100. Gilles C, Polette M, Piette J, Munaut C, Thompson EW,Birembaut P, Foidart JM. High level of MT-MMP expressionis associated with invasiveness of cervical cancer cells. Int JCancer 1996; 65:209–213.

101. Yamamoto M, Mohanam S, Sawaya R, Fuller GN, Seiki M,Sato H, Gokaslan ZL, Liotta LA, Nicolson GL, Rao JS. Differ-ential expression of membrane-type matrix metalloprotei-nase and its correlation with gelatinase A activation inhuman malignant brain tumors in vivo and in vitro. CancerRes 1996; 56:384–392.

102. Ohtani H, Motohashi H, Sato H, Seiki M, Nagura H. Dualover-expression pattern of membrane-type metalloprotei-nase-1 in cancer and stromal cells in human gastrointestinalcarcinoma revealed by in situ hybridization and immunoelec-tron microscopy. Int J Cancer 1996; 68:565–570.

103. Ellenrieder V, Alber B, Lacher U, Hendler SF, Menke A,Boeck W, Wagner M, Wilda M, Friess H, Buchler M, Adler G,

318 Ohkubo

Gress TM. Role of MT-MMPs and MMP-2 in pancreatic cancerprogression. Int J Cancer 2000; 85:14–20.

104. Ohuchi E, Imai K, Fujii Y, Sato H, Seiki M, Okada Y.Membrane type 1 matrix metalloproteinase digests intersti-tial collagens and other extracellular matrix macromolecules.J Biol Chem 1997; 272:2446–2451.

105. d’Ortho MP, Will H, Atkinson S, Butler G, Messent A,Gavrilovic J, Smith B, Timpl R, Zardi L, Murphy G.Membrane-type matrix metalloproteinases 1 and 2 exhibitbroad-spectrum proteolytic capacities comparable to manymatrix metalloproteinases. Eur J Biochem 1997; 250:751–757.

106. Fosang AJ, Last K, Fujii Y, Seiki M, Okada Y. Membrane-type 1 MMP (MMP-14) cleaves at three sites in the aggrecaninterglobular domain. FEBS Lett 1998; 430:186–190.

107. Rawlings ND, O’Brien E, Barrett AJ. MEROPS: the proteasedatabase. Nucleic Acids Res 2002; 30:343–346.

108. Southan C. A genomic perspective on human proteases asdrug targets. Drug Discov Today 2001; 6:681–688.


8

Mapping Intracellular ProteinNetworks

ZHAOZHONG HAN, ECE KARATAN, andBRIAN K. KAY

Argonne National Laboratory, BiosciencesDivision, Argonne, Illinois, U.S.A.

I. INTRODUCTION

In recent years, it has become well recognized that eukaryoticcells utilize protein–protein interactions for a number of

Abbreviations: Carboxy-terminal (C-terminal), complementary DNA(cDNA), enzyme-linked immunoabsorbant assay (ELISA), epithelial amilor-ide-sensitive sodium channel (ENaC), Eps15 homology (EH) domain, thehigh-mobility group protein 1 (HMGB1), glutathione-S-transferase (GST),growth factor-receptor binding protein (Grb), inhibitory concentration for50% activity (IC50), phosphotyrosine (pY), phosphotyrosine binding domain(PTB), PSD95-Discs large-ZO1 (PDZ) domain, Src homology (SH) 2 and 3domains, tumor necrosis factor b (TNF-b).

321

cellular processes, such as assembly of the cytoskeleton,

transference of signals during signal transduction, and tocompartmentalize proteins. Protein interaction modules often

serve to mediate these protein–protein interactions, and havethe following properties: they are typically 60–140 amino

acids in length, fold autonomously within the context of theprotein that contains the module, and bind short (i.e., 4–7

amino acids) segments of other proteins. Examples includethe Eps15 homology (EH) domain, the phosphotyrosine bind-

ing domain (PTB), the postsynaptic density=disc-large=ZO1(PDZ) domain, the Src homology (SH) 2 and 3 domains, and

the WW domain (Fig. 1).

Figure 1 Three-dimensional structure of peptides complexed withseveral protein interaction modules. Short peptides (stick diagrams)are shown complexed with modules (surface view). (A) STNPFLpeptide complexed with the middle EH domain of Eps15 (PDBaccession number 1FF1), (B) RALPPLPRY peptide complexed withthe SH3 domain of Src (1RLP), (C) AQTSV peptide complexed withthird PDZ domain of PSD-95 (1BE9), (D) GTPPPPYTVG peptidecomplexed with the WW domain of YAP (1JMQ), (E) pYEEI peptidecomplexed with the Src SH2 domain (1SPS), and (F) HIIENPN-pYFSDA peptide complexed with the PTB domain of Shc (1SHC).(pY represents phosphotyrosine.) For clarity sake, only shorterversions (underlined) of the peptides are shown in some casescomplexed to the domains.

322 Han et al.

Since protein interaction modules bind short peptidesequences, a fruitful approach to identifying their optimalpeptide ligands has been to screen combinatorial peptidelibraries. Such libraries can be synthesized on a solid sup-port either with mixtures of amino acids (1) or by the‘‘split–recombine’’ method (2). Alternatively, phage-displayedcombinatorial peptide libraries can be screened with a mod-ule and its ligand preferences can be deduced from theprimary structures of the selected peptides (3). Thus, byeither approach, one can deduce the optimal ligand prefer-ences of a protein interaction module in a few weeks’ time.Not only is this information useful in understanding howthe specificity of modules varies from one another, there isoften excellent correspondence between the primary struc-tures of the peptide ligands and regions within knowninteracting proteins. We have termed this phenomenon‘‘convergent evolution’’ (4). Hence, a productive process formapping protein–protein interactions within a proteome isto identify the optimal peptide ligands for a protein interac-tion module, predict the interacting proteins by computer,and test those hypothetical interactions that make intuitivebiological sense. Recent efforts in this direction are summar-ized below.

II. DOMAIN-MEDIATED INTERACTIONS

II.A. EH Domains

One domain, present in a number of proteins involved in thetransport and sorting of molecules within the cell, is the EHdomain (5). This domain is 100 amino acids in length andseveral three-dimensional structures have been determinedby nuclear magnetic resonance spectroscopy (6–8). To exam-ine the molecular recognition properties of this domain, aphage-displayed combinatorial 9-mer peptide library wasscreened by affinity selection with a glutathione-S-transferase (GST) fusion to the three EH domains presentin the endocytic protein, Eps15. Interestingly, every one ofthe selected peptides contained the tripeptide motif NPF (9).

Mapping Intracellular Protein Networks 323

This motif was demonstrated to be biologically relevant inseveral ways: (a) it occurred multiple times in Eps15 interact-ing proteins, (b) mutation of any of the NPF residuesdestroyed binding to the EH domain, and (c) the NPF residuescontact the most conserved residues on the surface of the EHdomain (10).

Phage display has been used to determine the specificityof EH domains present in other proteins (11,12). In each case,the EH domains selected peptides with the NPF motif,although certain residues predominated in the flankingregions, suggesting that the sequence context of the motifcontributes to the interaction. For instance, the binding ofphage-displayed peptide to the EH domains in intersectin isenhanced if the NPF motif is conformationally constrainedby flanking cysteines residues that form intramolecular disul-fide bonds (12). While the elucidation of the EH domain’sligand preferences have been proven important in mappingthe interaction of EH domain-containing proteins, the motifoccurs too frequently in proteomes to predict a manageablenumber of candidate interacting proteins for testing.

II.B. SH3 Domains

Src Homology 3 (SH3) domains are roughly 60 amino acids inlength and present in a wide array of membrane-associatedand cytoskeletal proteins, as well as proteins with enzymaticactivity and adaptor proteins without catalytic activity (13).At the time of this review, SH3 domains have been identifiedin 960 of the proteins in the proteomes of baker’s yeast,nematode, fruit fly, mustard plant, mouse, and man. In gen-eral, SH3 domains bind short (i.e., �7 amino acid) peptides,usually proline-rich, with the core motif of PxxP. Phage dis-play has been proven invaluable in determining the peptideligand specificity of SH3 domains, and predicting the cellularinteractions of a large number of proteins.

For example, the N-terminal SH3 domain of the adaptorprotein, Crk, selected peptides from a phage-displayed librarywith the motif, PxLPxK (Fig. 2). Searching GenBank with themotif demonstrated that this motif is present in Abl (PLLPTK),

324 Han et al.

C3G (PALPPK), clone S12 (PGLPSK), DOCK 180 (PPLPLK),and Eps15 (PALPPK), all of which were previously identifiedas Crk-interacting proteins, through a series of biochemicalexperiments (i.e., immunoprecipitation, far western blotting,and screening cDNA expression libraries). Therefore, severalof the cellular ligands of the Crk SH3 domain could have beenpredicted directly through phage display experiments.

The knowledge about specificity of individual SH3domains, determined through phage display, provides usefulinsight into how this module mediates specific protein–proteininteractions in the cell. For example, the motif that was selected

Figure 2 Selection of peptide ligands to the central SH3 domain ofCrk. A GST fusion to the N-terminal SH3 domain of human Crk wasused to screen an x6PxxPx6 combinatorial peptide library, displayedat the N-terminus of mature protein III of bacteriophage M13. Theprimary structures of the selected peptides are shown aligned tohighlight the consensus motif, PxLPxK. Through a combination ofcoimmunoprecipitation, filter lift, pull-down, and yeast two-hybridexperiments, the Crk SH3 domain has been shown to bind to Abl,C3G, Clone S12, DOCK, and Eps15.


for binding to the SH3 domain of Src (RxLPxLP) occurs inthe potassium channel Kv1.5 (RPLPxxP) and synapsin(RPQPPPP), while the motif selected for the SH3 domainof cortactin (þPPxPxKPxWL) occurs in CBP90 (KPPV-PPKPKMK) and Shank (KPPVPPKPKLK) (14). All of thesemolecular interactions have been shown to be SH3 domain-mediated (15–18), and one can verify the importance of thematching peptide sequences through deletion and mutagenesisexperiments. Inaddition,onecanusethesemotifs topredict thatthe type E neuronal Ca2þ channel alpha 1 subunit (RQLPPVP)and the faciogenital dysplasia protein (KPQVPPKPSYL) likelyinteract with the SH3 domains of Src and cortactin, respec-tively.Phagedisplay experiments have also been proven invalu-able in mapping the SH3 domain-mediated interactions ofendophilin and amphiphysin with synaptojanin (19), eps8 withAbi-1 (20), and Abp1p with the Ser=Thr kinases Prk1p andArk1p (21).

In recent years, the availability of sequenced genomeshas rendered large-scale proteome-wide analysis of protein–protein interactions plausible. It is now possible to constructprotein interaction networks consisting of all of the proteininteraction modules and their interaction partners in the pro-teome of an organism. Recently, Tong et al. (22) applied phagedisplay technology in a proteome-wide approach to create aninteraction network of all the SH3 domain-containing pro-teins and their binding partners in Saccharomyces cerevisiae.The authors identified 28 SH3 domains in the yeast proteomeand constructed GST fusions to each of the domains, 24 ofwhich could be expressed as soluble proteins in Escherichiacoli. After three rounds of selection, ligands were identifiedfor all but four of the SH3 domains, which the authors suspectmay not bind short peptides with sufficient (i.e., micromolar)affinity. For the remaining 20 SH3 domains, consensussequences were identified and used to scan the yeast pro-teome by computer for potential binding partners: the searchresulted in 206 proteins with 394 potential interactions. Theresulting set is displayed in Fig. 3 as a network consistingof nodes and lines, representing the individual proteins andinteractions between the proteins, respectively.

326 Han et al.

Figure 3 Interactions among yeast proteins predicted to bemediated by SH3 domains. (A) Diagram showing all the interac-tions predicted by phage display experiments. The nodes and linesrepresent proteins and protein–protein interactions, respectively.(B) Enlargement of those proteins with six SH3 domain-mediatedinteractions. Panels A and B are modified from Ref. 22. (C) Venndiagram comparing the SH3 domain-mediated interactions pre-dicted by phage display (light grey) with yeast two-hybrid screening(dark grey). Fifty-nine of the interactions predicted by phage dis-play (394) and yeast two-hybrid screening (233) overlap.


To confirm the SH3 domain-mediated interactions pre-dicted through phage display, Tong et al. (22) have utilizedyeast two-hybrid screens (23,24). Eighteen of the SH3 domainswere used as baits in two-hybrid screens of either individualopen reading frames or libraries of fragmented cDNAs fusedto the Gal4 activation domain, resulting in the identificationof 145 interacting proteins with 233 potential interactions(22). Comparison of the protein–protein interactions predictedby phage display and yeast two-hybrid analyses yielded 39proteins with 59 common interactions. To test the physiologi-cal relevance of the overlapping data sets, Las17p, a yeast pro-tein involved in actin assembly (25), was selected for furtherscrutiny. The two data sets predicted that 10 different SH3domain-containing proteins bind Las17p, of which three wereknown binding partners, four have been identified in othertwo-hybrid screens, and three are previously unidentifiedbinding partners. Coimmunoprecipitation analysis of yeastcells, which were transformed with hemagglutinin epitope-tagged Las17p and c-myc epitope-tagged SH3 domain-contain-ing proteins, demonstrated that many of the predicted proteininteractions occurred in vivo. To map the sites of interactionwithin the Las17p protein, five proline-rich regions of Las17pwere displayed separately as fusions to capsid protein D of bac-teriophage lambda and assayed for binding to the SH3domains in an enzyme-linked immunoabsorbant assay(ELISA). The experimental binding sites agreed in 9 out of10 cases with the binding sites inferred from the peptideligands selected by phage display.

II.C. PDZ Domains

PSD95-Discs large-ZO1 domains are 80–90 amino acids inlength, and are frequently found in proteins associated withthe cellular membrane, where they coordinate the assemblyof transmembrane and cytosolic components into multipro-tein complexes (26). The PDZ domain has an overall structurevery much like the PTB domain, even though they are unre-lated in function. It contains a core of five or six b-sheets(b1–6) and two a-helices (a1 and 2), and peptide ligands fit

328 Han et al.

into a hydrophobic pocket formed by a2, b2, and the conservedGLGF loop that connects the b1 and b2 strands. Unlike otherprotein interaction modules, PDZ domains recognize the freecarboxyl termini of target proteins, along with the penulti-mate four or five residues. Depending on the consensussequences of preferred ligands, PDZ domains have beengrouped into several classes: type I and II domains recognizeC-terminal peptides with the consensus sequence (T=S)x(L=V=I)-COOH or (F=Y) x(L=V=I) -COOH, respectively.

To learn what the �2000 PDZ domains in sequencedgenomes bind to, a useful approach has been to affinity selectligands from combinatorial peptide libraries and use them topredict the candidate interacting proteins. PDZ domain speci-ficity has been studied extensively with repertoires of chemi-cally synthesized combinatorial peptides (27) and a varietyof display systems. By fusing coding regions to the C-terminusof the Lac repressor (28), it has been possible to define the spe-cificity of the PDZ domain in neuronal nitric oxide synthase(29). Guided by the DxV-COOH consensus in the selected pep-tides, several candidate nNOS interacting proteins have beenidentified, including the glutamate and melatonin receptors.With a library displaying combinatorial peptides at the car-boxy terminus of the capsid D protein of bacteriophage l, Vac-caro et al. (30) have defined the ligand specificity of the sevenPDZ domains of the human INADL protein. From theconsensus sequences of the selected peptides, six of the PDZdomains correspond to either type I or II, with the seventhbelonging to a new type characterized by the presence of anacidic residue at the carboxyl-terminal position of the peptideligands.

While the peptide carboxylate group makes an importantcontribution to many PDZ–ligand interactions, it does not forall. When an M13 phage library of 12-mer combinatorial pep-tides displayed at the N-terminus of protein III was affinityselected with the PDZ domain of a-syntrophin, a componentof the dystrophin protein complex, three peptides withintramolecular disulfides were recovered (31). The peptidesshared residues with several known cellular ligands forthe a-syntrophin PDZ domain, but without a C-terminal


carboxylate group, suggesting that conformationally con-strained peptide ligands could mimic ligands for certainPDZ domains. The significance of this observation becameclearer with the subsequent discovery that b-hairpin peptide‘‘fingers’’ could fit into the groove of PDZ domains that nor-mally binds the C-termini of cellular ligands (32).

Recently, it has been discovered that it is possible todisplay peptides at the C-termini of proteins III or VIII ofbacteriophage M13 (33). In a modified phagemid display vec-tor, C-terminal fusions result in display levels comparable tothose achieved with conventional N-terminal fusions to eithercapsid protein. With a library of combinatorial peptides fusedto the C-terminus of protein VIII, peptide ligands havebeen selected for a variety of PDZ domains. Affinity selectionexperiments with the PDZ2 and PDZ3 domains of MAGI 3, amembrane-associated guanylate kinase, selected peptideswith the consensus Cys=Val-Ser=Thr-Trp-Val-COOH, whilethe PDZ3 domain selected only a single 7-mer peptidesequence (i.e., TRWWFDI-COOH). When synthetic peptidescorresponding to the PDZ2 consensus were tested for bindingto the domain, they were found to bind stronger than apeptide corresponding to the C-terminal sequence (i.e.,HTQITKV-COOH) of a natural PDZ2 ligand, the tumorsuppressor PTEN=MMAC (33); examination of the peptidessuggested that the Trp residue (underlined) in the Cys=Val-Ser=Thr-Trp-Val-COOH sequence contributed to thestronger binding. Affinity selection experiments of a libraryof combinatorial peptides displayed at the C-terminus ofpVIII with the PDZ domain of ERBIN, a basolateral proteinpredicted by yeast two-hybrid to interact with the mamma-lian ERBB2=HER2 receptor, yielded type 1 ligands thatbound strongly and specifically (34). The selected peptides clo-sely resembled the C-terminal sequence of three p120-likecatenins (d-catenin, ARVCF, and p0071). To test the signifi-cance of these predicted interactions with the ERBIN PDZdomain, both d-catenin and ARVCF were confirmed tointeract in vitro and in vivo, suggesting ERBIN may play arole of integrating cytoskeletal functions with epithelial cellmorphology and polarization.

330 Han et al.

II.D. WW Domains

Another protein interaction module to discuss is the WWdomain, named for the presence of two highly conserved tryp-tophan residues present in the 40 amino acid module (35). Todate, WW domains have been discovered in over 30 proteinsand often occur multiple times in a single protein, rangingfrom one to four copies. Characterization of two proteins,WBP1 and WBP2, discovered in a screen of a l-cDNA expres-sion library with a radiolabelled form of the WW domain ofYes-associated protein (YAP), revealed a common PPPPYsequence (36). In addition, a 10-mer peptide encompassingthis motif was able to bind the WW domain of YAP, and ala-nine scanning of this sequence revealed that the motif PPxY,where x is any amino acid, was necessary for this interaction.Based on this motif, a variety of proteins have beenproposed and confirmed to interact with WW domain-con-taining proteins, such as the cytosolic domain of epithelialamiloride-sensitive sodium channels (ENaC). Genetic evi-dence supports the biological importance of the interactionof the ENaC with an Nedd4, a ubiquitin ligase with threeWW domains, as mutations in the PPPY motif in the chan-nel cause Liddle’s syndrome (37), a form of hypertensionthat is caused due to the extended cellular half-life of thechannel (38).

The peptide ligand specificities of a variety of WWdomains have been defined with phage-displayed combinator-ial peptides, and this information has been used to predicttheir cellular interaction partners. Selection of peptides withthe YAP WW domain yielded the core sequence PPPPYP (39),which is present in the interacting proteins WBP1 and WBP2.Computer analysis also identified this motif within the p53binding protein-2 (p53BP2), and this protein–protein interac-tion was later confirmed in a yeast two-hybrid screen (40).Another matching protein was the p45 subunit of theNF-E2 transcription factor, which was confirmed by biochem-ical (41) and mutational (42) experiments. Screens of phagelibraries with a dozen other WW domains, related in primarystructure to the YAP WW domain, revealed an almost


absolutely conserved core motif of Pro-Xxx-Tyr among thepeptides selected, with different WW domains preferringvarying residues N- and C-terminal of this core (43). Compu-ter-aided searches of the protein database revealed severalcandidate interactions, including the Homologous to the E6-AP C-Terminus (HECT) domain, which is involved in theubiquitination of proteins.

Sometimes screens of phage-displayed combinatorialpeptide libraries with certain WW domains failed to yield pep-tide motifs. While screens with the WW domain of the cytos-keletal protein, utrophin, were initially negative, a largerprotein segment (including flanking EF hands and a ZZdomain) yielded binding phage that displayed the motif PPxY(44). This motif is present near the C-terminus of beta-dystro-glycan, a membrane protein known to bind utrophin and linkthe actin cytoskeleton to the extracellular basal lamina. Thus,the ability of the utrophin WW domain to bind peptides andcellular ligands is dependent on its structural context. Sup-port for this conclusion comes from biochemical and struc-tural analyses of the WW domain in a related protein,dystrophin: the WW domain of dystrophin requires its flank-ing EF hands to bind to a C-terminal segment of beta-dys-troglycan (45), and the beta-dystroglycan peptide binds acomposite surface formed by the WW domain and its flank-ing regions (46). Thus, when phage library screens are nega-tive, larger segments of the target proteins may benecessary to promote folding or provide additional contactsfor peptide ligand binding. Screens with the WW domainof the Pin1 protein also failed to yield peptide ligands (43).This negative result is to be expected, as the Pin1 WWdomain recognizes a posttranslationally modified peptidemotif, pSer =pThr-Pro, where pSer and pThr represent phos-phoserine and phosphothreonine, respectively (47).

II.E. PTB and SH2 Domains

Another protein interaction module is the PTB domain, a 100amino acid long domain that binds to specific phosphotyrosine

332 Han et al.

peptide sequences within proteins. In general, the specificityof PTB domains is mediated by amino acids N-terminal tothe phosphotyrosine in the cellular ligand (48). While chemi-cally synthesized peptide libraries have been quite useful inmapping the ligand preferences of individual PTB domains(49), it has also been possible to use phage-displayed peptidelibraries after incubating them with a protein tyrosine kinase,and then affinity selecting phage with the PTB domain. Forexample, incubation of phage displaying combinatorial9-mer peptides with the protein tyrosine kinase, Fyn, per-mitted the selection of phage that bound to the PTB domainof the adaptor protein Shc (50). Among the 18 different iso-lates, there was the consensus (F=Y)xNPTpYxx(Y=W), wherex and pY corresponds to any amino acid and phosphotyrosine,respectively. This phage-derived motif compares favorablywith the motifs (F=Y)xNPxpY and NPxpY, which were identi-fied from screens with synthetic combinatorial peptidelibraries (51) and inspection of the primary structures ofknown cellular ligands of the PTB domain of Shc, respectively.

Combinatorial peptide libraries have also played a usefulrole in mapping the specificity of the Src Homology 2 (SH2)domain, another module that binds to specific phosphotyrosinepeptide sequences within proteins. Unlike PTB domains, thespecificity of SH2 domains is mediated by amino acidsC-terminal to the phosphotyrosine in the cellular ligand. Theligand preferences of many SH2 domains have been definedwith synthetic peptides (49) and phage-displayed combinator-ial peptide libraries preincubated with protein tyrosinekinases. For example, from experiments in which phage wereincubated with a mixture of protein tyrosine kinases (c-Src,Blk, and Syk) and selected with the SH2 domain of growth fac-tor-receptor binding protein 2 (Grb2), it was found that thisSH2 domain recognizes the peptide sequence pY(M=E) NW(52). By a similar approach, the peptide ligand preferences ofthe SH2 domains of Grb2, Shc, Sli, and Rai have also beendefined (50).

From phage display of complementary DNA (cDNA)fragments, it is also possible to identify candidate interactingproteins for SH2 domains. For example, a library of phage


displaying fragments of leukocyte cDNA was incubated withprotein tyrosine kinase, Fyn, and selected with the tandemSH2 domains of the SHP-2, a cytoplasmic tyrosine phospha-tase. This approach led to the selection of clones encodingthe cytoplasmic domain of PECAM-1, a known cellular ligandfor SHP-2 (53). As more phage-displayed (i.e., lambda, T7)cDNA libraries become available, it should be possible toidentify cellular ligands of a large variety of different PTBand SH2 domains by this approach.

Even though SH2 domains typically bind to specific phos-photyrosine containing peptide sequences within cellular pro-teins, they have been also reported to bind proteins in aphosphotyrosine-independent manner. Two research groupshave selected nonphosphorylated peptide ligands fromphage-displayed combinatorial peptide libraries that can bindto Grb2 SH2 domain. One group isolated a single peptidesequence, CELYENVGMYC, that bound selectively to theSH2 domain of Grb2, and when synthesized in a cyclized, dis-ulfide form competed the binding of natural phosphopeptideligands, with an IC50 value of 15.5 micromolar (54). Replace-ment of either the tyrosine or asparagine residues in thispeptide led to a loss of binding, demonstrating that the YxNmotif contributes to binding to the Grb2 SH2 domain, evenwithout tyrosine phosphorylation. A second research groupalso screened three phage-displayed combinatorial peptidelibraries with a GST fusion to the Grb2 SH2 domain and iso-lated peptides that contained the YxN motif (55). While theselected peptides were preceded and followed by chargedand hydrophobic residues, respectively, only a subset of pep-tides contained pairs of cysteines residues. Synthetic formsof the peptides were competent to compete the recognitionof the phosphorylated epidermal growth factor (EGF-R) bythe Grb2 SH2 domain with an IC50 of 2–60 micromolar. Inter-estingly, synthesis of the phage-selected YxN peptides with aphosphotyrosine in place of the tyrosine residue increasedtheir affinity 10–100-fold for the Grb2 SH2 domain (55). Thus,the primary and secondary structural context of the YxNmotif contributes significantly in positioning peptides intothe SH2 domain pocket.

334 Han et al.

Nonphosphorylated peptide ligands have also beenselected for two other SH2 domains. The C-terminal SH2domain within the 85 kDa subunit (p85) of the phosphatidyli-nositol 3-kinase (PI 3-kinase) has been used to affinity selectpeptides from a phage-displayed combinatorial peptide library(56). The selected peptides shared the motif (L=A) A(R=K) IR,and a search of GenBank matched the serine=threonine kinaseA-Raf, which contained several examples of the motif. Coim-munoprecipitation experiments confirmed that these two pro-teins indeed form a protein complex in a mouse cell line. TheSH2 domain of Grb7, an adaptor-type signaling protein, wasused to screen a x4Cx10Cx4 combinatorial peptide library andselected peptides with the motif Y(A=D=E=G) N in the central10-mer region (57). From a second library randomizing thismotif, the investigators isolated peptides that bound the SH2domain of Grb7, and not that of Grb2 or Grb14. Synthetic,cyclized forms of these evolved peptides were able to competethe binding of Grb7 to the phosphorylated form of the ErbB3receptor, with an IC50 of �20 micromolar. While the sequencesof such peptides do not exactly match the sequences of Grb7interacting proteins, these peptides may prove useful in futuredrug discovery efforts (see below).

II.F. Chromo Shadow Domain

In chromatin, the heterochromatin-associated protein 1 (HP1)is thought to regulate heterochromatin structure throughinteractions with other proteins. To define the basis for selec-tive protein interactions, the C-terminal chromo shadowdomain of the Drosophila melanogaster HP1 protein was usedto select peptide ligands from a phage-displayed combinator-ial peptide library. Many of the peptides contained theconsensus sequence P-R=W=Y-V-L=M=V-L=M=V (58). Inter-estingly, not only is this pentapeptide sequence present inthe primary structure of many reported HP1-associated pro-teins but it is also occurs in the shadow domain, suggestingthat HP1 dimerization may occur through binding interac-tions with this motif as well. This conclusion is supportedby the observation that synthetic forms of the peptides bindto the shadow domain and also disrupt HP1 dimerization.


III. NONDOMAIN MEDIATEDPROTEIN–PROTEIN INTERACTIONS

In a number of cases, screens of phage-displayed combinator-ial peptide libraries have also been proven useful in mappingand predicting protein–protein interactions. Peptides selectedfor binding to protein VanR, the two-component signal trans-duction response regulator that controls expression of vanco-mycin resistance in Enterococcus faecium, yielded a 12-merpeptide that could be aligned with an 18-mer sequence ofthe catalytic center dimerization domain of VanS, a sequencewith which VanR also normally interacts (59). Amino acidreplacement experiments support a model in which theselected peptide mimics the VanS phosphorylatable sequencewith which the regulatory domain of VanR interacts, and thusfunctions as a ‘‘minimalist’’ analog of VanS. Selection of pep-tides that bind to tumor necrosis factor b (TNF-b) yielded apeptide sequence, RKEMGGGGGPGWSENLFQ, which con-tained two short motifs (RKEM, WSENLFQ) that matchedtwo regions spaced 25 amino acids apart in the primary struc-ture of tumor necrosis factor receptor 1 (TNFR1) (60). Selec-tion of peptides that bind to the two HMG boxes of thehigh-mobility group protein 1 (HMGB1) yielded a largecollection of diverse sequences, several of which matchedregions within a number of nuclear proteins and transcriptionfactors (61). Glutathione-S-transferase fusion proteinpull-down experiments confirmed that a number of thecandidate proteins could interact with HMGB1 in vitro.

IV. SOFTWARE FOR IDENTIFYINGCANDIDATE INTERACTING PARTNERS

Several resources are available for predicting interacting pro-teins in proteome databases based on matching the motif ofpeptide ligands identified through phage display. A position-specific scoring can be generated on-line (http:==blocks.fhcrc.org=blocks=process_blocks.html), in which the amino acidtolerance and frequency can be calculated at each positionof phage-displayed motif, and the BLOCKS file then used

336 Han et al.

for a BLAST search (62). With the SeqIt protein databasemining tool, available using the web-based program at ArrayGenetics (http:==www.arraygenetics.com), one can locate tar-get sequence elements within a protein database. Conversely,with the Scansite algorithm (http:==scansite.mit.edu), onecan identify protein sequences likely to bind to certain proteininteraction modules or likely to be phosphorylated by specificprotein kinases (63). For SH3 domains, the SH3-SPOTalgorithm can be used to predict cellular ligands (64).

With the increasing frequency of publications of particu-lar protein–protein interactions, a number of protein interac-tion databases have been created. The Molecular Interaction(MINT) database (http:==cbm.bio.uniroma2.it=mint=) in addi-tion to cataloging interactions, also stores other types of func-tional interactions, including enzymatic modifications of one ofthe partners (65). Presently, MINT contains 4568 interactions.The Database of Interacting Proteins (DIP) is a database(http:==dip.doe-mbi.ucla.edu) that documents experimentallydetermined protein–protein interactions. The DIP presentlycatalogs �11,000 interactions among 5900 proteins from> 80 organisms (66). The Biomolecular Interaction NetworkDatabase (BIND; http:==www.bind.ca=index.phtml) recordsinteractions between two objects (i.e., protein, DNA, RNA,ligand, molecular complex). The database also describes thecellular location, experimental conditions used to observe theinteraction, conserved sequence, molecular location, chemicalaction, kinetics, thermodynamics, and chemical state (67).BIND presently describes 6171 different interactions.

V. ANALYZING PREDICTED INTERACTIONS

Once the ligand preferences for a protein interaction moduleor a target protein have been defined, candidate interactingproteins can be tested for their ability to interact with themodule (or target protein) by a variety of methods such asaffinity chromatography (68), cross-linking, filter lifts (69),mass spectrometry (70), and two-hybrid screening (71) experi-ments. However, for a number of reasons, it is reasonable toexpect that some of the interactions predicted from the phage


display data may not actually occur in the cell or organism.First, it is possible that the two proteins are not expressedin the same cell type or at the same time, or occur in differentcellular compartments. Second, it is possible that due to pre-dominance of one protein–protein interaction, other protein–protein interactions occur infrequently in the cell. Third, itis also possible that another motif, not identified throughphage display, is responsible for molecular interactions.Fourth, it is also possible that peptide ligand sequence isnot accessible within the candidate protein either due to beingburied within the tertiary structure or the presence ofanother protein that sterically interferes with binding.Fifth, flanking resides (or posttranslational modifications)may influence the strength of binding of the candidate proteinwith the target. Thus, while phage display has the potentialto predict both physiologically relevant and nonrelevantinteractions, there are many experimental options availablefor investigators to sort through them quickly.

It should also be noted that since most phage libraries aretoo small to sample all 7-mer peptide permutations and not allcombinations are likely to be displayed efficiently (72,73), theremay be a slight ‘‘mismatch’’ between a phage display result andthe actual region within a protein that interacts with the targetprotein. To minimize such discrepancies, one can synthesizelarge numbers of peptides corresponding to all the regionswithin a proteome that resemble the phage display motif(19,74), monitor the actual binding of the peptides to the targetin vitro, and then rank order the list of candidate interactingproteins for further testing based on their strength of binding.In this way, the whole proteome can be assessed systematicallyfor its potential to interact with a particular target.

VI. RELEVANCE TO BIOTECHNOLOGY ANDDRUG DISCOVERY

Once peptide ligands have been discovered and used topredict protein interactions, one can consider using the pep-tides in three different manners for drug discovery. First,

338 Han et al.

the peptide can be injected, overexpressed, or linked to amembrane translocating sequence for delivery into cells(75), where it binds to the target protein and prevents aparticular protein–protein interaction. If the cell requiresthat interaction for a critical cellular function, then poten-tially there will be an observable, biological consequence.There are several precedents in the literature for this typeof perturbation experiment. For example, injection of syn-thetic peptide ligands for the SH3 domain of Src into frogoocytes leads to an acceleration of the completion of meiosis(76), whereas electroporation of peptide ligands for the SH3domain of Lyn blocks mast cell activation (77). These resultssuggest that Src and Lyn play important regulatory roles inoocyte maturation and mast cell activation, respectively.Recently, peptides with an LxxLL motif have been selectedfrom phage display that bind estrogen receptor a or b whencomplexed with estradiol (78). As these peptides appear tomimic an LxxLL sequence region in transcriptional coactiva-tors, they block transcriptional activity of either receptorwhen overexpressed in mammalian tissue culture cells(79,80). Finally, peptides that bind to and inhibit the activesites of enzymes have been overexpressed in bacteria to deter-mine whether particular enzymes are essential for growth(81,82). Thus, peptide ligands have great utility in validatingtargets as being suitable for drug discovery.

Second, once a peptide ligand with antagonist activityhas been identified, it can serve as a starting point in thedesign of a peptidomimetic inhibitor. The Kd’s of peptidesrecovered by phage display, when chemically synthesizedand tested in solution, range from 10mM to 500 nM (83).To enhance their affinity, it has been necessary to replacethe amino acids in the peptide with natural and unnaturalamino acids, as well as with different chemical entities. Inthis manner, the affinity of a peptide ligand for the CrkSH3 domain has been increased 20–100-fold (84,85). Thedesign of an effective peptidomimetic can also be greatlyaided by structure-based drug design, as demonstrated bythe conversion of a phage-displayed peptide ligand forthe major histocompatibility HLA-DR molecule (86) into a


protease resistant molecule that is 2000-fold more effec-tive than the parent heptapeptide in inhibiting T-cellproliferation (87).

Third, the phage-displayed peptides can be used to estab-lish a displacement assay in which collections of chemicalsand natural products can be screened for antagonists. Sincethe peptides frequently bind at sites of protein–protein inter-actions of a target, any molecule that prevents binding of thepeptide to the target presumably binds at the same siteand would function as an inhibitor (88). Therefore, syntheticforms of the peptides can be used to configure a competitive-binding assay for detecting chemical inhibitors. Many differ-ent types of such assays can be formatted in microtiter platewells and monitored by measuring changes in well color,luminescence, fluorescence, fluorescence polarization,time-resolved fluorescence, or fluorescence resonance energytransfer (89,90). As these assays are amenable to roboticsand automation, large chemical and natural product librariescan be screened for chemical modifiers of target activity, someof which may be developed into drugs.

REFERENCES

1. Songyang Z. Prog Biophys Mol Biol 1999; 71:359–372.

2. Lam K, Salmon S, Hersh E, Hruby V, Kazmierski W, Knapp R.Nature (London) 1991; 354:82–84.

3. Sparks A, Adey N, Cwirla S, Kay B. Kay BK, Winter J,McCafferty J, eds. Phage Display of Peptides and Proteins: ALaboratory Manual. San Diego: Academic Press, 1996.

4. Kay BK, Kasanov J, Knight S, Kurakin A. FEBS Lett 2000;480:55–62.

5. Confalonieri S, Di Fiore PP. FEBS Lett 2002; 513:24–29.

6. de Beer T, Carter RE, Lobel-Rice KE, Sorkin A, Overduin M.Science 1998; 28:1357–1360.

7. Koshiba S, Kigawa T, Iwahara J, Kikuchi A, Yokoyama S.FEBS Lett 1999; 442:138–142.

340 Han et al.

8. Whitehead B, Tessari M, Carotenuto A, van Bergen enHenegouwen PM, Vuister GW. Biochemistry 1999; 38:11,271–11,277.

9. Salcini AE, Confalonieri S, Doria M, Santolini E, Tassi E,Minenkova O, Cesareni G, Pelicci PG, Di Fiore PP. GenesDev 1997; 11:2239–2249.

10. de Beer T, Hoofnagle AN, Enmon JL, Bowers RC, YamabhaiM, Kay BK, Overduin M. Nat Struct Biol 2000; 7:1018–1022.

11. Paoluzi S, Castagnoli L, Lauro I, Salcini AE, Coda L, Fre S,Confalonieri S, Pelicci PG, Di Fiore PP, Cesareni G. EMBO J1998; 17:6541–6550.

12. Yamabhai M, Hoffman NG, Hardison NL, McPherson PS,Castagnoli L, Cesareni G, Kay BK. J Biol Chem 1998; 273:31401–31406.

13. Macias MJ, Wiesner S, Sudol M. FEBS Lett 2002; 513:30–37.

14. Sparks A, Rider J, Hoffman N, Fowlkes D, Quilliam L, Kay B.Proc Natl Acad Sci USA 1996; 93:1540–1544.

15. Holmes TC, Fadool DA, Ren R, Levitan IB. Science 1996;274:2089–2091.

16. Foster-Barber A, Bishop JM. Proc Natl Acad Sci USA 1998;95:4673–4677.

17. Ohoka Y, Takai Y. Genes Cells 1998; 3:603–612.

18. Naisbitt S, Kim E, Tu JC, Xiao B, Sala C, Valtschanoff J,Weinberg RJ, Worley PF, Sheng M. Neuron 1999; 23:569.

19. Cestra G, Castagnoli L, Dente L, Minenkova O, Petrelli A,Migone N, Hoffmuller U, Schneider-Mergener J, Cesareni G.J Biol Chem 1999; 274:32001–32007.

20. Mongiov AM, Romano PR, Panni S, Mendoza M, Wong WT,Musacchio A, Cesareni G, Paolo Di Fiore P. EMBO J 1999;18:5300–5309.

21. Fazi B, Cope MJ, Douangamath A, Ferracuti S, Schirwitz K,Zucconi A, Drubin DG, Wilmanns M, Cesareni G, CastagnoliL. J Biol Chem 2002; 277:5290–5298.

22. Tong AH, Drees B, Nardelli G, Bader GD, Brannetti B,Castagnoli L, Evangelista M, Ferracuti S, Nelson B, Paoluzi S,


Quondam M, Zucconi A, Hogue CW, Fields S, Boone C, CesareniG. Science 2002; 295:321–324.

23. Fields S, Song O. Nature (London) 1989; 340:245–246.

24. Phizicky E, Fields S. Microbiol Rev 1995; 59:94–123.

25. Naqvi SN, Zahn R, Mitchell DA, Stevenson BJ, Munn AL.Curr Biol 1998; 8:959–962.

26. Harris BZ, Lim WA. J Cell Sci 2001; 114:3219–3231.

27. Songyang Z, Fanning AS, Fu C, Xu J, Marfatia SM,Chishti AH, Crompton A, Chan AC, Anderson JM, CantleyLC. Science 1997; 275:73–77.

28. Cull M, Miller J, Schatz P. Proc Natl Acad Sci USA 1992;89:1865–1869.

29. Stricker NL, Christopherson KS, Yi BA, Schatz PJ, Raab RW,Dawes G, Bassett DE Jr, Bredt DS, Li M. Nat Biotech 1997;15:336–342.

30. Vaccaro P, Brannetti B, Montecchi-Palazzi L, Philipp S,Citterich MH, Cesareni G, Dente L. J Biol Chem 2001; 276:42122–42130.

31. Gee SH, Sekely SA, Lombardo C, Kurakin A, Froehner SC,Kay BK. J Biol Chem 1998; 273:21980–21987.

32. Hillier BJ, Christopherson KS, Prehoda KE, Bredt DS, LimWA. Science 1999; 284:812–815.

33. Fuh G, Pisabarro MT, Li Y, Quan C, Lasky LA, Sidhu SS.J Biol Chem 2000; 275:21486–21491.

34. Laura RP, Witt AS, Held HA, Gerstner R, Deshayes K,Koehler MF, Kosik KS, Sidhu SS, Lasky LA. J Biol Chem2002; 277:12906–12914.

35. Ilsley JL, Sudol M, Winder SJ. Cell Signal 2002; 14:183–189.

36. Chen HI, Sudol M. Proc Natl Acad Sci USA 1995; 92:7819–7823.

37. Tamura H, Schild L, Enomoto N, Matsui N, Marumo F,Rossier BC. J Clin Invest 1996; 97:1780–1784.

342 Han et al.

38. Staub O, Abriel H, Plant P, Ishikawa T, Kanelis V, Saleki R,Horisberger JD, Schild L, Rotin D. Kidney Int 2000; 57:809–815.

39. Linn H, Ermekova KS, Rentschler S, Sparks AB, Kay BK,Sudol M. Biol Chem 1997; 378:531–537.

40. Espanel X, Sudol M. J Biol Chem 2001; 276:14514–14523.

41. Gavva NR, Gavva R, Ermekova K, Sudol M, Shen CJ. J BiolChem 1997; 272:24105–24108.

42. Mosser EA, Kasanov JD, Forsberg EC, Kay BK, Ney PA,Bresnick EH. Biochemistry 1998; 37:13686–13695.

43. Kasanov J, Pirozzi G, Uveges AJ, Kay BK. Chem Biol 2001;8:231–241.

44. Tommasi di Vignano A, Di Zenzo G, Sudol M, Cesareni G,Dente L. FEBS Lett 2000; 471:229–234.

45. Rentschler S, Linn H, Deininger K, Bedford MT, Espanel X,Sudol M. Biol Chem 1999; 380:431–442.

46. Huang X, Poy F, Zhang R, Joachimiak A, Sudol M, Eck MJ.Nat Struct Biol 2000; 7:634–638.

47. Verdecia MA, Bowman ME, Lu KP, Hunter T, Noel JP. NatStruct Biol 2000; 7:639–643.

48. Yan KS, Kuti M, Zhou MM. FEBS Lett 2002; 513:67–70.

49. Songyang Z, Shoelson SE, McGlade J, Olivier P, Pawson T,Bustelo XR, Barbacid M, Sabe H, Hanafusa H, Yi T, et al.Mol Cell Biol 1994; 14:2777–2785.

50. Dente L, Vetriani C, Zucconi A, Pelicci G, Lanfrancone L,Pelicci PG, Cesareni G. J Mol Biol 1997; 269:694–703.

51. Songyang Z, Margolis B, Chaudhuri M, Shoelson SE, CantleyLC. J Biol Chem 1995; 270:14863–14866.

52. Gram H, Schmitz R, Zuber JF, Baumann G. Eur J Biochem1997; 246:633–637.

53. Cochrane D, Webster C, Masih G, McCafferty J. J Mol Biol2000; 297:89–97.


54. Oligino L, Lung FD, Sastry L, Bigelow J, Cao T, Curran M,Burke TR Jr, Wang S, Krag D, Roller PP, King CR. J BiolChem 1997; 272:29046–29052.

55. Hart CP, Martin JE, Reed MA, Keval AA, Pustelnik MJ,Northrop JP, Patel DV, Grove JR. Cell Signal 1999; 11:453–464.

56. King TR, Fang Y, Mahon ES, Anderson DH. J Biol Chem 2000;275:36450–36456.

57. Pero SC, Oligino L, Daly RJ, Soden AL, Liu C, Roller PP, Li P,Krag DN. J Biol Chem 2002; 277:11918–11926.

58. Smothers JF, Henikoff S. . Curr Biol 2000; 10:27–30.

59. Ulijasz AT, Kay BK, Weisblum B. Biochemistry 2000;39:11417–11424.

60. Pillutla RC, Hsiao Kc K, Brissette R, Eder PS, Giordano T,Fletcher PW, Lennick M, Blume AJ, Goldstein NI. BMCBiotechnol 2001; 1:6.

61. Dintilhac A, Bernues J. J Biol Chem 2002; 277:7021–7028

62. Smothers JF, Henikoff S. Comb Chem High ThroughputScreen 2001; 4:585–591.

63. Yaffe MB, Leparc GG, Lai J, Obata T, Volinia S, Cantley LC.Nat Biotechnol 2001; 19:348–353.

64. Brannetti B, Via A, Cestra G, Cesareni G, Citterich MH. J MolBiol 2000; 298:313–328.

65. Zanzoni A, Montecchi-Palazzi L, Quondam M, Ausiello G,Helmer-Citterich M, Cesareni G. FEBS Lett 2002; 513:135–140.

66. Xenarios I, Salwinski L, Duan XJ, Higney P, Kim SM,Eisenberg D. Nucl Acids Res 2002; 30:302–305.

67. Bader GD, Hogue CW. Bioinformatics 2000; 16:465–477.

68. Gavin AC, Bosche M, Krause R, Grandi P, Marzioch M, BauerA, Schultz J, Rick JM, Michon AM, Cruciat CM, Remor M,Hofert C, Schelder M, Brajenovic M, Ruffner H, Merino A,Klein K, Hudak M, Dickson D, Rudi T, Gnau V, Bauch A,Bastuck S, Huhse B, Leutwein C, Heurtier MA, Copley RR,Edelmann A, Querfurth E, Rybin V, Drewes G, Raida M,

344 Han et al.

Bouwmeester T, Bork P, Seraphin B, Kuster B, Neubauer G,Superti-Furga G. Nature (London) 2002; 415:141–147.

69. Kurakin A, Bredesen D. J Biomol Struct Dyn 2002; 19:1015–1030.

70. Deshaies RJ, Seol JH, McDonald WH, Cope G, Lyapina S,Shevchenko A, Verma R, Yates JR III. Mol Cell Proteomics2002; 1:3–10.

71. Uetz P. Curr Opin Chem Biol 2002; 6:57–62.

72. Peters EA, Schatz PJ, Johnson SS, Dower WJ. J Bact 1994;176:4296–4305.

73. Rodi DJ, Soares AS, Makowski L. J Mol Biol. 2002; 322:1039–1052.

74. Kramer A, Reineke U, Dong L, Hoffmann B, Hoffmuller U,Winkler D, Volkmer-Engert R, Schneider-Mergener J. J PeptRes 1999; 54:319–327.

75. Lindgren M, Hallbrink M, Prochiantz A, Langel U. TrendsPharmacol Sci 2000; 21:99–103.

76. Sparks AB, Quilliam LA, Thorn JM, Der CJ, Kay BK. J BiolChem 1994; 269:23853–23856.

77. Stauffer TP, Martenson CH, Rider JE, Kay BK, Meyer T.Biochemistry 1997; 36:9388–9394.

78. Paige LA, Christensen DJ, Gron H, Norris JD, Gottlin EB,Padilla KM, Chang CY, Ballas LM, Hamilton PT, McDonnellDP, Fowlkes DM. Proc Natl Acad Sci USA 1999; 96:3999–4004.

79. Chang C, Norris JD, Gron H, Paige LA, Hamilton PT, Kenan DJ,Fowlkes D, McDonnell DP. Mol Cell Biol 1999; 19:8226–8239.

80. Norris JD, Paige LA, Christensen DJ, Chang CY, HuacaniMR, Fan D, Hamilton PT, Fowlkes DM, McDonnell DP.Science 1999; 285:744–746.

81. Tao J, Wendler P, Connelly G, Lim A, Zhang J, King M, Li T,Silverman JA, Schimmel PR, Tally FP. Proc Natl Acad SciUSA 2000; 97:783–786.

82. Benson RE, Gottlin EB, Christensen DJ, Hamilton PT. 2003;47:2875–2881.


83. Hyde-DeRuyscher R, Paige LA, Christensen DJ, Hyde-DeRuyscher N, Lim A, Fredericks ZL, Kranz J, Gallant P,Zhang J, Rocklage SM, Fowlkes DM, Wendler PA, HamiltonPT. Chem Biol 2000; 7:17–25.

84. Posern G, Zheng J, Knudsen BS, Kardinal C, Muller KB, Voss J,Shishido T, Cowburn D, Cheng G, Wang B, Kruh GD, BurrellSK, Jacobson CA, Lenz DM, Zamborelli TJ, Adermann K, Hana-fusa H, Feller SM. Oncogene 1998; 16:1903–1912.

85. Nguyen JT, Turck CW, Cohen FE, Zuckermann RN, Lim WA.Science 1998; 282:2088–2092.

86. Hammer J, Valsasnini P, Tolba K, Bolin D, Higelin J, TakacsB, Sinigaglia F. Cell 1993; 74:197–203.

87. Bolin DR, Swain AL, Sarabu R, Berthel SJ, Gillespie P, HubyNJ, Makofske R, Orzechowski L, Perrotta A, Toth K, CooperJP, Jiang N, Falcioni F, Campbell R, Cox D, Gaizband D,Belunis CJ, Vidovic D, Ito K, Crowther R, Kammlott U, ZhangX, Palermo R, Weber D, Guenot J, Nagy Z, Olson GL. J MedChem 2000; 43:2135–2148.

88. Kay BK, Kurakin A, Hyde-DeRuyscher R. Drug Discov Today1998; 3:370–378.

89. Houston JG, Banks M. Curr Opin Biotech 1997; 8:734–740.

90. Grøn H, Hyde-DeRuyscher R. Curr Opin Drug Discov Dev2000; 3:636–645.

346 Han et al.

9

High Throughput and High ContentScreening Using Peptides

ROBERT O. CARLSON,ROBIN HYDE-DERUYSCHER, and

PAUL T. HAMILTON

Karo Bio USA, Inc., Durham,North Carolina, U.S.A.

I. INTRODUCTION

Phage-displayed peptide libraries can be used to isolatespecific, high-affinity peptides to almost any protein target.Peptides have been identified for a wide range of protein tar-gets including antibodies (1,2), enzymes (3,4), G proteins (5),nuclear receptors (6,7), cytokine receptors (8,9), cytokines (10),transcription factors (11), and protein interaction domains(12–14). The peptides identified by affinity selection fromphage-displayed peptide libraries bind to biologically relevantsites on these target proteins and therefore can serve as

347

valuable reagents for dissecting the biological function ofproteins as well as tools for drug discovery.

In this chapter, we will cover how peptides identified byphage display bind to functional sites on protein targets, suchas enzyme active sites, protein–protein interaction sites, andDNA binding sites. These peptides can, therefore, be used asmolecular probes of the protein target. For an enzyme, anactive site-directed peptide can be used to screen for enzymeinhibitors. For proteins that undergo conformational changesbased on activation by another protein or by ligand binding,the peptides can be used to monitor the conformationalstate of the target protein. In drug discovery research, thesepeptide characteristics can be exploited to develop rapid, invitro assays for high throughput compound screening andcharacterization.

II. PEPTIDES AS ENZYME INHIBITORS

II.A. Peptides Identified by Phage Display areDirected to Biologically Relevant Sites

The evidence available from a number of laboratories doingpeptide selections by phage display indicates that the pep-tides isolated with protein targets are directed to only oneor a few sites on the protein, and in most cases, the sites thepeptides bind are functional sites. These peptides, therefore,often modulate the biological activity of the target proteinand serve as ‘‘surrogate ligands’’ of the target.

A wide variety of enzyme classes have been used astargets for peptide selection by phage display (see review inRef. 4). Many of the peptide ligands identified by phage dis-play for enzymes bind at the active site of the enzyme andinhibit the enzymatic activity with Ki or IC50 values rangingfrom 60nM to 3mM. In a study of seven diverse enzymes,Hyde-DeRuyscher et al. (3) were able to isolate peptides thatbound to each of the seven enzyme targets. For six of theseven targets, they were able to isolate peptides that inhibitedthe enzymatic activity of the target enzyme. For example,hexokinase, which catalyzes the phosphorylation of glucoseby ATP to yield glucose-6-phosphate, was used as a target

348 Carlson et al.

for peptide selection and seven peptides were identified. Thetwo highest affinity peptides were synthesized and used forkinetic analysis. Both peptides were competitive inhibitorsof hexokinase with respect to glucose. The best inhibitor pep-tide had a Ki of 80 nM with respect to glucose and 100nM withrespect to ATP. In another example, the target enzyme wasalcohol dehydrogenase and again peptides were identifiedwhich were competitive inhibitors of alcohol dehydrogenasewith Ki values in the range of 100–800nM. For hexokinaseand alcohol dehydrogenase, the substrates for the enzymesare small molecular weight compounds with no peptidic orproteineous characteristics. Yet, peptide ligands were identi-fied that are competitive inhibitors of the enzymes with Ki

values in the hundred nanomolar range.In another study, Sperinde et al. (15) used phage display

to identify a peptide that bound and inhibited Dnase II. Whentested in activity assays, the peptide had a Ki of 2mM, butwas sparingly soluble in aqueous solution. Extending thesequence of the peptide to increase solubility improved theinhibition of Dnase II to give a Ki of 0.4 mM. These researcherswere able to use the Dnase inhibitor peptide to enhance thetransfection of DNA into various cell types.

The finding that peptides identified by phage displaybind to protein functional sites, and do not bind randomlyor nonspecifically to a protein’s surface, indicates that func-tional sites on proteins have a common set of physicochemicalfeatures that are recognized by the peptides. These features ofa functional site are the basis of ligand binding, whether theligand is an enzyme substrate, an interacting protein partner,or a phage-displayed peptide. A number of studies havelooked at proteins to determine the characteristics of a func-tional site. Binding sites are usually grooves or depressionsin the protein surface (16), often with exposed hydrophobicgroups. Mattos and Ringe (17,18) have used small organiccompounds as probes to define hydrophobic binding sites onthe surface of a protein. They found that functional sites have,in addition to the characteristics listed above, bound watermolecules that make specific interactions with polar groupsin the site. Ligand binding easily displaces the water

High Throughput High Content Screening Using Peptides 349

molecules in the functional site. Another characteristic offunctional sites appears to be flexibility of the residues thatmake up the site. DeLano et al. (19) examined a common bind-ing site in the hinge region of an Fc (crystallizable fragment)of immunoglobulin G that interacts with four different nat-ural proteins. In addition, they did phage-display selectionexperiments with peptide libraries and found high-affinitypeptides that bound at this ‘‘consensus’’ binding site on theFc. Crystallographic analysis of the ‘‘consensus’’ binding sitewith the peptide showed that the peptide adopted a structurethat is different from the natural Fc binding proteins, despitethe fact that the interactions of the peptide with the bindingsite mimicked the interactions of the other interacting pro-teins. In particular, the binding site on the Fc fragmentunderwent conformational changes in order to complementthe surface residues of each binding partner.

The peptides, therefore, appear to be able to fit into agroove or depression on the protein surface and displace thebound water molecules. Hydrophobic residues on the peptideinteract with the exposed hydrophobic residues in the func-tional site. These two features probably provide the majorityof the energy for binding. Binding specificity may come fromcomplementarity of charge and polar residues between thepeptide and the binding site. In addition, the binding site isflexible and therefore can adapt its shape to accommodatethe peptide ligand.

II.B. Peptides as Surrogate Ligands in HighThroughput Screening Assays

Since the peptides bind at a functional site of a target protein,a compound that prevents binding of the peptide to the targetprotein would also likely bind at the same site on the targetand would therefore function as an inhibitor (3,20). The pep-tides, therefore, can be used as surrogate ligands to configuresimple, competitive-binding assays to detect inhibitors inhigh throughput screening (HTS). Peptide-based, competi-tive-binding assays have been configured in a variety ofdetection formats including radioactivity-based scintillation

350 Carlson et al.

proximity (SPA), luminescence, fluorescence polarization(FP), time-resolved fluorescence (TRF), and fluorescenceresonance energy transfer (FRET).

These formats vary as to whether the peptide or the pro-tein or both need to be labeled. In some formats, the protein isimmobilized on a solid support, in other formats, the peptideis immobilized, and in still other formats, both protein andpeptide are in solution. Regardless of the format, peptide-based, competitive binding assays can detect inhibitors. Forexample, a peptide-based assay was developed for tyrosyl-tRNA synthetase using a peptide identified by phage display(3). Using a variety of detection formats, a series of fourknown tyrosyl-tRNA inhibitors were tested in peptide-basedassays (3,21). In each assay format, the ability to detect theinhibitor and the potency of the compound were determinedand compared to the activity of the compounds in a biochem-ical assay (Table 1). All four of the inhibitors were detected ineach of the assay formats and the potencies for the compoundsin the peptide-based assays were comparable with the poten-cies of the compounds in the enzyme activity assay. Thepeptide-based assays were able to detect compounds withpotencies ranging from nM to mM.

II.C. HTS Example: Deoxyxylulose-PhosphateReductoisomerase (DXR)

Isoprenoids are a group of compounds found in all livingorganisms. In bacteria, isoprenoids are synthesized fromisopentenyl diphosphate (IPP) via a mevalonate-independentpathway. IPP is essential for bacterial growth, therefore,enzymes, such as DXR that are involved in the biosynthesisof IPP, are valid targets for antibacterial drug discovery (22).For many targets in drug discovery, HTS assays are basedon biochemical activity assays. For DXR, the activity assayinvolves organic separation of radioactive substrates and pro-ducts, which is a process that is not very amenable to an HTSassay. We developed a peptide-based, HTS assay for DXR andused it to identify compounds that inhibited the enzymaticactivity of DXR (23).


Table 1 Comparison of Detection Methods for Peptide-Based Assays

Assay format

InhibitorNPC0101 IC50

(mM)


(mM)


(mM)


(mM)

Enzymatic activity 0.04 0.20 3.0 1.6Target immobilized—SPA 0.03 0.19 7.0 18Target immobilized—TRF 0.01 0.12 6.0 10Peptide immobilized—TRF 0.02 0.24 3.7 10Fluorescence polarization (FP) 0.02 0.22 2.1 3.6Fluorescence resonanceenergy transfer (FRET)

0.03 0.45 6.8 12

352

Carlso

net

al.

The dxr gene sequence was amplified by PCR from E. coligenomic DNA, cloned into an expression vector with a biotiny-lation recognition sequence, expressed and biotinylated in E.coli, and purified. Purified, biotinylated DXR was immobilizedon streptavidin-coated 96-well microtiter plates and 20 differ-ent phage-displayed peptide libraries were used to selectDXR-specific peptides by methods described previously (3).A series of DXR-specific peptides were identified and usedto build a TRF assay. The DXR peptide-based assay was usedto screen a collection of 32,000 structurally diverse com-pounds, from which 111 hits were obtained for a 0.35% hitrate (Table 2). Each of these 111 compounds specificallyblocked the binding of the DXR-specific peptide to DXR. Fromtesting of the 89most potent inhibitors in theDXR-peptide com-petition assay, 30 of the compounds (34%)were found to beDXRenzyme inhibitors with IC50 values < 20mM, making themcandidates for further investigation for potential applicationas antibacterial agents.

The peptide-based, surrogate ligand assay for DXRproved to be a rapid, reliable method to identify enzyme inhi-bitors in a high throughput screen of a compound collection.In addition to our internal efforts with DXR and other targets,other groups (24) have successfully used peptide-based, surro-gate ligand assays to identify enzyme inhibitors. Since thepeptides are directed to functional sites on the target protein,this method can be used to develop HTS assays for targets ofunknown biochemical function. Surrogate ligand assays fortargets of unknown function can identify compounds that

Table 2 Compound Screening Results for DXR

32,000 compounds were screenedScreening statistics: CV¼ 11%; Z0 ¼ 0.6111 actives identified using DXR-specific peptide-based assay (0.35%) in aTRF assay format

30 of 89 compounds tested in a functional assay inhibit the enzymaticactivity of DXR with potencies < 20 mM

Potency of inhibitors ranged from 0.3 to 20 mM


can be tested directly for effects on the disease model. Inaddition, peptide-based assays can be used to develop HTS-compatible assays for targets where traditional biochemicalassays are difficult or impossible to configure and run in ahigh throughput format.

II.D. Peptides as Tools for Target Validation

The observation that the phage-display process selectspeptides that bind at functional sites of target proteins ratherthan to random sites on the protein surface can be used todevelop tools for target validation. The basic approachinvolves isolation of peptide ligands to a target protein usingphage display followed by expression of the peptide inside thecell and monitoring for phenotypic effects. Peptides that bindat functional sites on the target will block that target functioninside the cell and produce a change in phenotype. For targetsthat are essential for cell growth, the phenotype would begrowth inhibition or cell death. This ‘‘protein knockout’’ tech-nique has been applied successfully to several essential E. coliproteins (25,26). Tao et al. (25) used phage display to selecta peptide that bound to prolyl-tRNA synthetase, fused thepeptide to GST, and demonstrated growth inhibition of thebacterial cells upon specific expression of the peptide-GSTfusion. Similarly, Benson et al. (26) took a set of six targetsinvolved in a wide range of basic bacterial cellular processes[DNA replication (DnaN), transcription (RpoD), DNA super-coiling (GyrA), lipid A biosynthesis (LpxA), protein secretion(SecA), and one target of unknown function (Era)] and identi-fied target-specific peptides by phage display. When the tar-get-specific peptides were expressed inside E. coli cells asGST fusions, the cells stopped growing. Overexpression ofthe target protein relieved the peptide-induce growth arrest,indicating that the peptides were inhibiting an essentialfunction required for bacterial growth. For bacterial cells, thismethod validates the protein as a suitable target for antibac-terial drug discovery, but it also validates the peptide astargeting an essential functional site. The peptide–proteinpair can then be used to develop an HTS-compatible assay

354 Carlson et al.

for antibacterial drug discovery as described above for DXR.This approach does not depend on prior knowledge of thebiochemical activity or function of the target protein and there-fore may be well suited for use with targets of unknownfunction that are identified by genomic and genetic techniques.

III. PEPTIDES AS CONFORMATIONAL PROBES

Another useful aspect of peptides identified using phagedisplay is their ability to bind to targets in a conformationallysensitive fashion. This property can be used to developspecific probes useful for many drug discovery applications.We have used conformationally sensitive peptides to buildassays for G-protein coupled receptors and nuclear receptors.

III.A. G Protein Coupled Receptors

G-protein coupled receptors (GPCRs) are seven transmembranereceptor proteins that are expressed in almost all human tissuesand play a fundamental role in signal transduction and physiol-ogy. GPCRs, therefore, are potential therapeutic targets fortreating many diseases. Of the 483 marketed drugs in 1996,greater than 40% were directed at GPCR targets (27). In2000, 26 of the top 100 pharmaceutical products were therapeu-tics thatmodulateGPCR function (28). This accounts for sales ofover 23.5 billion USD and represents approximately 9% of thetotal global pharmaceutical sales (29).

The human genome has been estimated to consist of35,000 genes of which approximately 750 are GPCRs. Almosthalf of these sequences are likely to encode sensory receptors.The remaining 400 or so receptors, however, represent poten-tial targets for drug discovery, of which only about 30 are tar-gets of existing therapeutics. In addition, an estimated 160 ofthese receptors are ‘‘orphan’’ receptors for which there is noknown ligand or function (29).

Most assays for GPCRs rely on monitoring levels ofintracellular second messengers such as inositol trispho-sphate or calcium. The problem with this approach is that


many different cellular functions can effect the production ofthese second messengers, leading to a high rate of inter-ference from compounds that have no effect on the activityof the GPCR in question. In addition, these assays are carriedout in cells, requiring a large commitment of resources in cellculture to perform the assays. Our approach was to simplifythe system by doing two things (30). First, remove the assayfrom the cellular environment and utilize isolated membranesthat contain all of the components necessary to monitor GPCRsignaling. This includes the receptor itself and the hetero-trimeric G protein, consisting of the alpha, beta, and gammasubunits. Second, we wanted to develop probes that wouldmonitor GPCR activation by measuring the activation stateof the G-alpha protein associated with the receptor. GPCRsare ligand-dependent guanine nucleotide exchange factorsfor G proteins. When a ligand binds to the receptor, it causesa signal to be transduced through the membrane from theGPCR to the associated G protein. This causes the G-alphasubunit to undergo a conformational shift whereby it releasesGDP and binds GTP. This results in a change in contactsbetween the G-alpha and beta–gamma subunits of theheterotrimeric G protein, which then interact with othereffector proteins, initiating the signal transduction cascade.Thus, the basic function of the receptor is to cause a ligand-dependent conformational change in the G-alpha subunit.

To identify peptides with the needed specificities, we pro-duced recombinant G-alpha subunit in E. coli and used this asa target for phage display. For this set of experiments, we pre-sented the G-alpha protein in either the inactive, GDP boundstate or the activated GTP bound state. As shown in Fig. 1, wewere able to identify peptides that bound with very distinctnucleotide specificities. For example, one series of peptides,designated T-specific peptides, bound with high affinity toG-alpha protein with GTP bound. Another class of peptidesbound only to G protein complexed with GDP, designatedD-specific peptides. Yet, a third class of peptides bound tothe G-alpha protein in a nucleotide independent fashion; thatis they bound to the G-alpha protein in both the GTP and theGDP states.

356 Carlson et al.

In addition, the peptides showed remarkable specificityfor different subtypes of G proteins. For instance, peptidesthat bound to G-alpha I did not bind to G-alpha S and simi-larly, peptides that bound G-alpha S did not bind G-alpha I.We have been able to identify peptides that are not onlyspecific for the activation state of the G protein but also thesubtype of G protein.

These peptides are very specific probes for the activationstate of the G-alpha protein, providing a tool for the directreadout of the activation state of the receptor. As shown

Figure 1 Binding specificities of the peptides identified usingphage display. Synthetic peptides were conjugated with streptavi-din-alkaline phosphatase and used in an ELISA format to probeincreasing concentrations of Ga protein immobilized on a microtiterplate. (A) A peptide identified using Ga protein loaded with GTPbinds specifically to GTP loaded Ga. (B) A peptide identified usingGa protein loaded with GDP binds specifically to GDP loaded Ga.(C) A peptide identified using Ga protein loaded with GDP bindsto Ga independent of nucleotide loading.


in Fig. 2, these peptides can be used to format assays thatmonitor receptor activation. If T-specific peptides are linkedto a reporter molecule, they can be used to format a simplebinding assay or a homogeneous FRET assay. These assayscombine a number of features that are desirable in a HTSplatform. They monitor only functional activation of thereceptor, not merely a binding event and so are an efficientway to find agonists for the receptor. The assay uses semipur-ified components and so cell culture is not required on a dailybasis to set up the assay. The assays are routinely performedin volumes of 30 mL in 384 well plates; small amounts of mate-rials are required for each assay and the FRET-based assaydoes not require washing. All of these properties make thisan ideal platform for automation.

As shown in Fig. 3, an assay of this type was used tomonitor the activity of the beta 2 adrenergic receptor. Theagonist isoproterenol was titrated into the assay and showeda dose-dependent activity with an EC50 value that isconsistent with literature values. In addition, this activitycould be antagonized with the inverse agonist ICI 118551(Fig. 3). This indicates that the assay monitors activation

Figure 2 Schematic representation of detecting activation of areceptor using a peptide that binds to the activated form of Ga.

358 Carlson et al.

through the receptor and not nonspecific changes in this sys-tem. We also have formatted assays for G-alpha I coupledreceptors, such as the M2 muscarinic acetylcholine receptor.Figure 3 shows that the agonist carbachol also elicits adose-dependent increase in activity when using this assay.The signal could also be reduced by the addition of the antago-nist atropine.

Using the G-alpha I and the G-alpha S systems, wewere able to demonstrate that there is specific coupling

Figure 3 Assays of (A) B2 adrenergic receptor (Gas coupled) and(B) muscarinic acetylcholine receptor M2 (Gai coupled). Membranesfrom SF9 cells infected with baculovirus expressing the receptor,and the heterotrimeric G proteins were used with a peptide specificfor activated Ga to monitor the activation of the receptor. Increas-ing concentrations of the agonist was added to the assay. Increasingsignal from bound probe is an indicator of receptor activation(graphs on left). Membranes were treated with agonist at 80% fullactivation and increasing concentrations of antagonist were added(graphs on right).


through each receptor. In addition, this system exhibits theexpected activation profile for partial agonists. Several partialagonists as well as a full agonist were used for the beta 2 adre-nergic receptor. In each case, the potency observed for eachcompound was consistent with published values and theirefficacy also mirrored the known activation profile for thesecompounds. Thus, the assay can also provide reasonablepharmacological data on the compounds in question.

The peptide probes used in these GPCR assays are speci-fic for the G-alpha subunit. G proteins associate with manydifferent GPCRs. This system, therefore, can be used with avariety of GPCRs by changing the receptor that is expressedin the cells. For example, we obtained the D-1 dopaminereceptor and tested it using our G-alpha S system. Toassess the activity of the receptor, we carried out a series ofdose–response experiments with the agonist dopamine andevaluated the reproducibility over several experiments.We were able to obtain reproducible signal-to-backgroundratios for dopamine of 5:1 with no additional optimization.This highlights the modular nature of this GPCR assaysystem.

As a preliminary assessment of the ability of the systemto be used as an effective tool to screen large collections ofcompounds, and to address the cross talk or interference bycompounds which would normally activate many pathwaysand cellular assays, we used a collection of 640 known phar-macologically active compounds. This set included 11 knownagonists to the M2 acetylcholine receptor. All 11 agonistswere detected by the assay and only one ‘‘new’’ compoundwas determined to be an agonist. Upon further testing, wefound that the activity of the ‘‘new’’ compound could bereversed by the inverse agonist for this receptor, indicating itwas a specific agonist. This ‘‘new’’ compound, TMB8, has beendescribed previously as an allosteric modulator of the M2receptor, but not an agonist (31). Therefore, TMB8 may nothave been previously identified as a M2 receptor agonist, sincea direct binding assay would also not detect this compound,which further demonstrates the utility of this approach to findnovel compounds that modulate GPCR function.

360 Carlson et al.

To summarize for GPCRs, we have identified peptidesthat can be used to monitor the activation state of receptorsin crude membrane preparations by assessing the nucleotidebinding state of the G-alpha protein. This assay is functional,does not rely on live cells, and the data obtained reflects thepharmacological properties of individual receptors.

III.B. Nuclear Hormone Receptor

III.B.1. Complexity in Nuclear Receptor Signaling

Nuclear receptors are a group of transcription factors thatregulate expression of target genes in response to ligand bind-ing. The signal of ligand binding to nuclear hormone receptoris transduced through induced changes in receptor surfaceconformation that alter interactions with coregulatory pro-teins. These coregulatory proteins in turn regulate cellularfunctions through a variety of mechanisms, which includebut are not exclusively limited to transcriptional regulation(32,33). Different ligands induce distinct receptor surface con-formations leading to differences in the type and affinity ofcoregulatory protein interactions. The nature of these inter-actions thereby defines the biological activity of the ligand.Since the expression of coregulatory proteins can be cell-typedependent, such differences in expression may be a factor inthe ability of nuclear receptor ligands to exert cell-typespecific effects, such as is observed for the selective estrogenreceptor modulators (SERMs) (34).

There are a wide variety of possible coregulatory proteins.Most identified to date interact primarily with nuclear recep-tor activation function 2 (AF2), which is located in the ligandbinding domain (LBD). These include p160 steroid coactiv-ator family members, p300 and related integrator proteins,TRAP=mediator complex, and various other coactivators (32).The primary mechanism for ligand-mediated regulationof AF2 interaction with coactivators has been defined throughmutagenesis and crystallography (35–38). Using the estrogenreceptor as an example, binding of the agonist 17-b-estradiolinduces receptor helices 3, 4, 5, and 12 to create a hydrophobic


‘‘coactivator groove’’ that in turn binds LXXLL motifs presentin coactivator proteins (39). ER antagonists, such as4-OH-tamoxifen, disrupt the position of helix 12, preventingcoactivator binding through LXXLL motifs (40).

The identity of residues flanking and filling out theLXXLL motif has also been shown to be critical in directingaffinity and specificity of binding to nuclear receptors. Usingphage display with a focused LXXLL peptide library, Changet al. identified a multitude of LXXLL peptides that differen-tially bound ERa, ERb, progesterone receptor (PR), gluco-fcorticoid receptor (GR), and androgen receptor (AR) in thepresence of their respective agonists. Recent studies of thebinding of LXXLL containing sequences from p160=SRCfamily proteins to ERa and ERb have also revealed significantdifferences in receptor and ligand specificity and affinity (41–43). Studying the four variant LXXLL motifs of SRC-1, Par-ker and colleagues demonstrated substantial differences inaffinity for ERa, AR, retinoid receptors (RARa and RXRa).They also defined a minimal core LXXLL motif of �2 to þ6in which a hydrophobic residue at �1 and a nonhydrophobicresidue at þ2 correlated with highest affinity binding. In theLXXLL motif of thyroid hormone receptor binding protein(TRBP), mutation of a single serine at the �3 position alteredTRBP binding to ERa and ERb, and increased binding tothyroid hormone receptor (TR) and RXR (44). Thus, not allLXXLL sequences are created equal with regard to ligandand=or nuclear receptor specificity, and hence with regardto biological function.

LXXLL motifs are not the only sequences involved inligand-dependent interaction of proteins with nuclear recep-tors. As demonstrated recently for PPARg, the corepressorSMRT also binds the AF2 coactivator groove, but throughan LXXXIXXXL motif (45). In contrast to coactivator binding,antagonist-mediated disruption of helix 12 leads to stabiliza-tion of binding for this corepressor motif. Ligand-dependentinteractions of corepressors with nuclear receptors also occuroutside of the AF2 domain (46). Androgen receptor hasbeen shown to interact through its amino terminus withcoregulators containing FXXLF or WXXLF motifs, in an

362 Carlson et al.

androgen dependent fashion (47). The coactivator NRIF3interacts specifically with TR or RXR through a domain thatcontains an LXXIL motif, but this motif alone does not deter-mine this specificity (48). Crystal structures of ERa withLXXLL containing peptide sequences from the p160 coactiva-tor TIF2 revealed that the box 3 sequence actually bindsthrough an LXXYL rather the adjacent LXXLL motif presentin the sequence (49).

There are also many proteins that do not containeither LXXLL or any of the other motifs described above,which exhibit ligand-dependent interaction with nuclearreceptors, based upon yeast two-hybrid, coimmunopre cipita-tion, or GST pull-down experiments. Table 3 contains a non-exhaustive list of these putative nuclear receptor interactingproteins that lack LXXLL or similar motifs. The proteins

Table 3 Nuclear Receptor Interacting Proteins that LackNR Box Motifs

Interactingprotein Activity

Interactingnuclearreceptor(s) Reference

BAG-1 Apoptosis inhibitor VDR, GR (53)C=EBP alpha Enhancer binding protein GR (54)CAPER Coactivator ER (55)Caveolin-1 Scaffold protein AR (56)CDC37 Chaperone AR (57)CIA Coactivator ER (58)DAP-3 Apoptosis mediator GR (59)GMEB-1=2 Coactivator GR (60,61)HBO1 Histone acetylase AR (62)MDM2 Ubiquitin ligase ER (63)Nucleolin Undefined GR (64)PAK6 Protein kinase ER, AR (65)PBX Homeodomain protein TRa (66)PNRC2 Coactivator SF1, ERR, other (67)REA Corepressor ER (68)Smad3 Signal transduction AR (69)SP3 Transcription factor ER (70)SRA RNA coactivator AR (71)Thioredoxin Reducing enzyme GR (72)


listed in Table 3 cover a diverse range of functions that arepotentially regulated through ligand-dependent changes innuclear receptor conformation. Some of these activities donot clearly have direct links to transcriptional regulation.Lacking LXXLL or related motifs, these proteins are likelyto bind outside of the coactivator groove. Of course, basedupon interaction studies alone, it is not clear that proteinsare binding directly to nuclear hormone receptors.

Thus, substantial evidence exists for complexity in ligand-mediated regulation of nuclear receptors with coregulatoryproteins. In the context of novel nuclear receptor liganddiscovery for therapeutic application, this complexity poses bothdifficulties and opportunities. The difficulties center on thedaunting task of creating a novel ligand screening processsufficient to cover the potential nuclear receptor signaling com-plexity. This task of course is simplified in proportion to the levelof understanding that exists for a given nuclear receptormechanism of action within a specific biological response ofinterest for drug discovery. However, despite a rich databaseof potential nuclear receptor signaling mechanisms, the actualmechanisms involved in specific physiological events in mostcases remains poorly defined. Therefore, the signaling complex-ity can create a burden in expanding the sphere ofmodel systemsthat may need to be included in nuclear receptor drug discoveryprograms. On the other hand, the realization of this complexityis at theheart of a recent resurgence in interest in drug discoveryfor nuclear receptors. The understanding that ligands can directchanges in nuclear receptor signaling pathways throughinduced conformational changes, and that these ligand directedevents are highly cell-type dependent, has led to the appreciationthat discovery of novel ligands with exquisitely selective activ-ities is a distinct possibility for all nuclear hormone receptors.

What follows in this section is a description of a technol-ogy, termed Molecular Braille� profiling, designed to detectligand-induced changes in nuclear hormone receptorconformation. The basic principle involves use of molecularprobeswith differential selectivity and affinity for binding siteson the nuclear receptor surface that are altered in response toligand binding.Work with estrogen receptor is presented as an

364 Carlson et al.

example of technology that can be applied to nuclear hormonereceptors in general, and in principle, to any other protein thatundergoes induced conformational changes.

III.B.2. Phage Display for Identifying Peptides toProbe Nuclear Receptor SurfaceConformation

We currently use peptides exclusively as molecular probes inthe Molecular Braille technology. Phage display provides atechnique for identifying peptides that have altered bindingaffinity for nuclear receptors in response to ligand binding.Such peptides can be used as tools to probe the differencesin conformation resulting from binding of different nuclearreceptor ligands. We have had very good success in usingphage display to find diverse peptide sequences that bind tonuclear receptors, including ERa, ERb, GR, TR, and AR. Agood example of the process is provided by a recent phage-dis-play experiment at Karo Bio, which was designed to identifypeptides that selectively bind the ERb ligand binding domaineither unliganded (apo), or complexed with its natural ago-nist, 17-b-estradiol (E2) or the SERM raloxifene. The phage-display approach for nuclear receptors is essentially identicalto the process described for enzymes (3,6). After the secondround of phage affinity selections, bound phage were eluted,and plated, from which about 1500 plaques in total wererandomly picked for amplification from the three selectionconditions. These amplified, clonal phage were then testedfor specific binding using a phage ELISA technique, andabout half were subsequently found to bind selectively tothe original target, relative to irrelevant protein or uncoatedplastic wells. Based upon signal intensity, a subset of over100 of those selectively bound phage were then retested inthe phage ELISA format for ligand selectivity using a panelof ER ligands bound to ERb-LBD, including the SERMstamoxifen and raloxifene, the antagonist faslodex, and the ago-nists diethylstilbestrol (DES) and genistein. The results of thisphage ELISA are depicted in Fig. 4. In general, the ligand selec-tivity was consistent with the selection conditions. For example,


phage selected against E2 bound ERb-LBD subsequently recog-nize that complex, and the other agonist complexes (DES andgenistein), but not the SERMs or antagonist. The conversewas found for phage selected against raloxifene bound ERb.Curiously, phage selected against unliganded ERb show higherbinding to the apo receptor as would be expected, but also haveeither the E2 or raloxifene selection pattern. The latter mayresult from these peptides inducing a receptor conformation

Figure 4 Ligand specificity of phage selected with ERb. A set of168 phage derived from affinity selections, using immobilized ERbwithout ligand, or saturated with estradiol or raloxifene, weretested for ligand specificity using a phage ELISA protocol. Biotiny-lated ERb LBD was to bound to streptavidin-coated 384 well plates,ligands [E2, diethylstilbesterol (DES), genistein (GEN), 4-OHtamoxifen (TAM), raloxifene (RAL), or faslodex (FAS)] were addedto 1 uM final concentration, together with phage. After 1 hr of incu-bation at room temperature, the plates were washed, and HRP-con-jugated anti-aM13 antibody was added, followed by another roundof washing, and then addition of HRP substrate ABTS to quantifybound antibody. The phage ELISA signal is based upon resultingabsorbance at 405nM. The resulting ligand specificity patternsare shown in relation to the original phage selection condition[ERb unliganded (APO), or with E2 or raloxifene].

366 Carlson et al.

similar to that induced by E2 or raloxifene, or unliganded ERmay spontaneously take on E2 or raloxifene conformations inthe absence of the ligands, which is then recognized by peptideswith affinity for these conformations.

In an effort to better understand the relative similarity ofthe isolated phage based upon this ligand selectivity phageELISA, we submitted the ELISA dataset to principal com-ponent analysis (PCA), using software from Spotfire. PCAprovides a way for reducing multivariable datasets to fewerdimensions that aid visualization and can help in understand-ing which variables are changing the most. The seven compo-nent phage binding patterns were reduced through PCA tothree dimensions, which are plotted in Fig. 5 for the phageanalyzed from all of the selection conditions. In that plot,each symbol represents the seven compound binding patternof a single phage, and the closer the spheres are in space,the more similar their binding patterns. The phage isolatedfrom using either E2 or raloxifene bound ERb-LBD clearlycluster separately on opposite sides of the plot, while theapo-ER selected phage spread throughout the plot. LXXLLmotif containing peptides were isolated using either apo orE2-bound receptor, and as expected all of those peptidesshow agonist conformation binding patterns. However, manynon-LXXLL peptides also show agonist conformation bindingpatterns.

Principal component analysis was also applied to thesame dataset, but for use in comparing the phage binding pat-terns relative to reference compound-bound (agonists DES,E2, and genistein; SERMs raloxifene and 4-OH-tamoxifen;antagonist faslodex) or unliganded receptor. In Fig. 6, eachsphere in the PCA plots represents the binding pattern ofmultiple phage for each reference compound (or apo receptor).In Fig. 6A, the phage binding pattern is based upon phagedisplaying the peptides b I, b III, a=b III, a=b IV, and a=b V,which we have previously used extensively as probes forligand-induced conformational changes for ER (6,50). Asexhibited by the proximity of the spheres, this set of pre-viously identified peptides can discriminate faslodex ortamoxifen-bound ERb-LBD distinctly from the agonist or


raloxifene-bound or the apo receptor. However, the agonistsDES, E2, and genistein are poorly discriminated, and raloxi-fene-bound ERb does not appear significantly different fromthe apo receptor. Based upon sequence and ligand-inducedbinding patterns from the recent E2, raloxifene, or apo phageaffinity selections as described above, 10 new phage were

Figure 5 Discriminating the ligand specificity patterns of ERbphage using PCA. The ligand specificity patterns based upon thephage ELISA data depicted in Fig. 4 were submitted to PCA usingstatistical analysis software from Spotfire. Each symbol in the threedimensional plot represents the complete ligand specificity patternfor each individual phage. The distance separating points in spaceis proportional to the similarity of the ligand specificity patterns.The shape of the symbol relates to the original selection condition:spheres¼ERbþE2; cubes¼ERb apo; pyramids¼ERbþ raloxifene;crosses¼phage from other ER selections. Amongst the spheres,white spheres represent NR box motif containing sequences, andgray spheres represent sequences without NR box motifs.

368 Carlson et al.

chosen. These new phage were added to the six previouslyidentified phage to evaluate the binding pattern withthe same set of reference conditions. With the new phageadded, the agonist-induced conformations appeared distinct,while the discrimination of the SERM, antagonist, and apoconformations was maintained (Fig. 6B). Thus, the objectiveto find peptides for better ER agonist-induced conformationappears to have been achieved through the new selections.This use of PCA provides a good method for determiningwhether the isolated phage have the potential for discriminat-ing the ligand-induced receptor conformations of interest.

Figure 6 Discriminating the peptide-binding pattern for ligand-induced conformations of ERb using PCA. The phage ELISA datadepicted in Figs. 4 and 5 was used to evaluate the extent to whichphage could discriminate between ERb conformations induced bydifferent ligands. In (A), the peptide-binding pattern is depictedfor ERb apo or bound with E2, genistein, DES, faslodex, raloxifene,or 4-OH tamoxifen, based upon the phage-displayed peptides, b I, bIII, a=b I, a=b III, a=b IV, and a=b V, which were previously used todiscriminate differences in ligand-induced ER conformation. In (B),the peptide-binding patterns for the same set of ligands are basedupon the peptides in (A), plus 10 additional peptides from theERb phage affinity selections described in Figs. 4 and 5. The dis-tance separating each sphere in space is proportional to the similar-ity of the peptide-binding patterns.


Using PCA to visualize individual phage binding patterns andalso phage binding patterns for specific receptor forms facili-tates selection of a subset of phage for synthesis to be appliedin assays for more precise detection of induced receptor con-formational changes.

III.B.3. Methods for Use of Peptides to DetectLigand-Induced Nuclear ReceptorConformational Changes

We utilize two general methods for detecting ligand-inducednuclear receptor conformational changes with peptides. Oneapproach is entirely in vitro, using purified receptor and syn-thetic peptides. The precise format used to detect the amountof receptor complexed with peptide in response to ligandbinding is quite flexible. We currently use TRF or FRET tomeasure the receptor–peptide complex. The TRF format issimilar to an immunosorbent assay, with formats that involvebinding either biotinylated synthetic peptide or biotinylatednuclear receptor to streptavidin-coated 384 well plates. Thenbiotinylated receptor (for peptide coated plates) or peptide (forreceptor coated plates) is added with or without ligand.Unbound receptor or peptide is washed away, and strepta-vidin–europium cryptate conjugate is added to detect boundreceptor or peptide through TRF (Fig. 7A). The FRET formatdoes not require immobilization of any assay component.Streptavidin–europium cryptate conjugate is bound with bio-tinylated, synthetic peptide and then mixed with receptorthat is complexed with allophycocyanin (APC), with or with-out ligand. Excitation at 340nm of the europium peptide tagleads to emission at 620nm, or if the peptide is receptor bound,the excitation energy can directly transfer to the receptor APCtag, which emits at 665nm. Fluorescence at 665nm due to340nm excitation is therefore directly proportional to amountof peptide–receptor complex (Fig. 7B).

Another method frequently used involves a mammaliantwo-hybrid protein–protein interaction assay,whichwe call Cel-lular Braille2 assay, in which peptide-Gal4 DNA-bindingdomain fusion and nuclear receptor VP16-fusion are transiently

370 Carlson et al.

Figure 7 Examples of screening formats in which peptides areused as probes of ligand-induced nuclear receptor conformationalchanges. (A) Immobilized: peptide on plate (POP)-TRF MolecularBraille assay in which biotinylated peptide is bound to a streptavi-din-coated plate. Test ligand and biotinylated receptor complexed toa streptavidin–europium cryptate conjugate are added, and quan-tity of receptor bound to peptide is measured through TRF fromthe europium. This assay may also be ‘‘turned over’’ into a targeton plate (TOP)-TRF Molecular Braille assay in which biotinylatedreceptor is bound to a streptavidin-coated plate and the biotinylatedpeptide is linked to the streptavidin–europium cryptate conjugate.(B) Homogeneous: FRET based Molecular Braille assay in whichthe peptide is complexed to streptavidin–europium and the receptoris complexed to streptavidin–allophycocyanin. The interactionbetween the peptide and receptor is measured by monitoring thesignal at 665nm. (C) Cell based: Cellular Braille assay in whichGAL4-peptide fusion, nuclear receptor-VP16, and GAL4 promoterdriven luciferase reporter are transiently transfected into cells.Effects of ligand on luciferase transcription are directly propor-tional to the extent of ligand-induced receptor interaction withpeptide.


expressed in cells along with a Gal4-driven reporter (Fig. 7C)(51). Gal4-peptide bound to VP16-receptor results in activationof the Gal4-driven reporter, which is therefore an indirect mea-sure of peptide–nuclear complex concentration. Changes in thereporter readout in response to ligand are thereby used to detectaltered nuclear receptor surface conformation.

The results obtained from these alternate methods aretypically equivalent. As shown in Fig. 8, the effects of E2 orthe SERM 4-OH tamoxifen on interaction of ERb-LBD witha=b I peptide show similar selectivity and affinity. The differ-ences are primarily empirical or practical. Routinely, a fewernumber of peptide sequences will be usable in the Cell-ular Braille assay. In particular, the Cellular Braille assayexcludes peptides that require intact disulfide bonds for bind-ing, because disulfide bonds are disrupted in the reducingenvironment of the cell. We also find fewer peptides providedetectable ligand-induced signals in the FRET vs. the TRFformat, but the precise reason for this is unclear. In general,the in vitro formats allow for much higher throughputcompared to the Cellular Braille assay. Perhaps the mostimportant reason for using the Cellular Braille assay is thata source of purified, unliganded receptor is not required. Forsome of the nuclear receptors, obtaining purified protein inquantity is difficult, and purification in the absence of ligandcan be impossible due to receptor instability in the absence ofligand. For those receptors, the Cellular Braille assay is themethod of choice. The Cellular Braille assay also allowssurface conformation detection in the natural environmentof the receptor, including the complexity of interactions withother intracellular proteins and with DNA. This ill-definedcomplexity, however, can also complicate the interpretationof results.

For ER, the Molecular Braille assay is currently ourmethod of choice, based upon the relative robustness, through-put, and well-defined nature of the assay. Previous publicationsdescribed theuse of the b I, b III, a=b I, a=b III, a=b IV, and a=bVpeptides in the TRF format to discriminate ligand-induced con-formational changes for ER (6,50). That set of peptides is usefulfor discriminating SERM-induced conformations, but not for

372 Carlson et al.

agonist-induced conformations (6,50), as depicted in Fig. 6A.When 10 peptides from the recent affinity selections usingapo and E2 and raloxifene-bound ERb-LBD were added to theprevious set of six peptides for the Molecular Braille assay,discrimination between E2, DES, and genistein-induced

Figure 8 Comparison of methods for measuring ligand-inducedpeptide interaction with nuclear receptor. Comparison of E2 or 4-OH tamoxifen-induced changes in interaction between ERb anda=b I peptide using (A) POP-TRF Molecular Braille assay, (B)TOP-TRF Molecular Braille assay, or (C) Cellular Braille assay asdescribed in Fig. 7.


conformations became apparent (Fig. 6B). This is consistentwith the characteristics of these peptides as displayed onphage (Fig. 4). Thus, the phage ELISA data served as a goodbasis for selecting peptides for synthesis, in that the ability todiscriminate the agonist conformations apparent with phagewas recapitulated with the synthetic peptides.

III.B.4. Role of Molecular Braille Technology inDrug Discovery for Nuclear Receptors

Drug discovery for nuclear receptors typically emphasizesoptimization of ligand binding affinity and selectivity. Deter-mination of relative agonistic and antagonistic activityfollows, most often based upon a cellular assay involving sometype of classical transactivation event mediated by a nuclearreceptor response element. This approach is appropriate forfinding novel agonists or antagonists that can mimic or blockendogenous agonist effects, respectively, with high affinity andpossibly nuclear receptor subtype selectivity, if desired.-However, this approach is not adequate for finding SERMs orsimilarly tissue-specific modulators for other nuclear receptors.

To address tissue specificity, both cell-based and animalmodels are employed. For example, one potential liability ofSERMs is agonistic activity in uterine epithelium, whichmay lead to uterine cancer (52). Models used to screen for thisactivity include measuring agonism in uterine endometrialcancer cells in culture or tracking changes in uterine weightin mice or rats. Each screen has drawbacks, but is acceptedas appropriate methods for filtering out SERMs with poten-tial for uterotrophic activity. Uterine cancer, however, is onlyone potential side effect arising from tissue-specific actions ofSERMs, but incorporating cell-based or animal models forevery tissue of interest within a drug discovery program isimpractical.

Molecular Braille technology currently provides the mostefficient and comprehensive method for evaluating the poten-tial of nuclear receptor ligands to affect biological responses ina tissue dependent manner. This is intrinsically based uponthe fact that ligand-induced conformational changes,

374 Carlson et al.

measured in the Molecular Braille assay, drive the coregula-tory protein interactions that mediate tissue-specific signal-ing of nuclear receptors. Furthermore, the peptides utilizedin the Molecular Braille assay may resemble actual sequencesin coregulatory proteins that similarly bind nuclear receptorsin response to ligand. Therefore, binding of these peptidesmay mimic physiologically relevant events, and thus couldprove useful for correlating ligand-induced changes in peptidebinding to ligand-induced biological responses. To the extentthat this is true, Molecular Braille technology offers a surro-gate, comprehensive method for assaying potential proteininteractions nuclear receptors may encounter in the myriadof tissues in which the receptors are active. These potentiallymimicked interactions may involve currently known andunknown coregulatory proteins. Consistent with the promi-nence of LXXLL motifs in ER agonist-induced protein interac-tions, our phage affinity selections using estradiol-bound ERprimarily yield peptides containing LXXLL or related motifs.These LXXLL motif-containing peptides interact with ERwith the same ligand specificity observed for interaction ofcoactivator proteins that contain LXXLL motifs. However,not all of the peptides found with estradiol bound ER, andnone of the peptides found for ER bound with SERMs, containLXXLL motifs. Yet, these peptides can bind with the affinityand ligand specificity seen with LXXLL motif containing pep-tides, and therefore it stands to reason that these non-LXXLLpeptides are also binding functionally relevant sites on ER.One example of this is the ability of the non-LXXLL peptidea=b III, but not the LXXLL peptide a=b I, to inhibit tamoxi-fen-induced, ER-dependent signaling through a C3 promoter(50). The mechanism of ER transactivation through this C3promoter remains undefined, yet it apparently can involve atamoxifen-induced recruitment of a coregulatory protein thatbinds through a non-LXXLL binding site on ER (50). Thisprovides an example of how peptides discovered throughphage display from our large Karo Bio combinatorial phagelibrary may represent sequences in coregulatory proteinsthat have yet to be discovered, but may be important for ERphysiological actions of interest. As such, Molecular Braille


technology may provide information that is unavailable forany other source.

Considering the central role ligand-induced conforma-tional changes play in directing nuclear signaling, efforts tocharacterize this activity are essential for novel liganddiscovery, and should be included as early as possible in thediscovery process. We envision use of the Molecular Brailleassay immediately downstream from in vitro ligand affinityand selectivity characterization. As such, ligands subse-quently tested in cell-based or animal models will have thedesired affinity and selectivity, and will also be associatedwith an induced conformation, as described by the MolecularBraille pattern or fingerprint, that can be correlated withdesirable or undesirable in vivo effects.

IV. SUMMARY

Peptides isolated from phage-display peptide libraries arepowerful tools for drug discovery. These peptides are directedto functional sites on target proteins and can be used to developassays for a wide range of different targets. For enzymes, thepeptides target the active site and therefore function as inhi-bitors of the enzyme. These active site-directed peptides canbe used as surrogate ligands for enzymes to develop HTSassays to screen chemical compound collections for smallmolecule inhibitors, such as was described in this chapterfor the enzyme Dxr.

For targets involved in signal transduction, we have beenable to isolate peptides that can specifically sense the confor-mation of the target protein allowing the development of uniqueassays for GPCRs and nuclear receptors. The assays for GPCRsare based on peptides that specifically bind to the active confor-mation of the G-alpha subunits of heterotrimeric G protein. Theassay is formatted as a cell-free, homogeneous assay thatdetects functional activation of the GPCR and does not rely onany downstream reporter.

In addition, we have developed peptide-based assays fornuclear receptors, Molecular Braille technology, which probe

376 Carlson et al.

the conformation of the nuclear receptors induced by ligandbinding. Often, the peptides identified by phage-displayaffinity selections on nuclear receptors mimic proteins thatinteract with nuclear receptors. For example, we have iden-tified peptides that contain the LXXLL motif found in nuclearreceptor coactivators such as SRC-1. Using a panel of nuclearreceptor-specific peptides, we can evaluate the receptorconformation induced by various ligands and classify theligands based on peptide-binding profile. Biological activityof a given nuclear receptor ligand is determined by theinduced receptor conformation, therefore, the peptide-bindingprofile should ultimately be able to predict the biologicalactivity of a ligand.

REFERENCES


2. Cortese R, Monaci P, Luzzago A, Santini C, Bartoli F, CorteseI, Fortugno P, Galfre G, Nicosia A, Felici F. Selection of biolo-gically active peptides by phage display of random peptidelibraries. Curr Opin Biotechnol 1996; 7:616–621.

3. Hyde-DeRuyscher R, Paige LA, Christensen DJ, Hyde-DeRuyscher N, Lim A, Fredericks ZL, Kranz J, Gallant P,Zhang J, Rocklage SM, Fowlkes DM, Wendler PA, HamiltonPT. Detection of small-molecule enzyme inhibitors with pep-tides isolated from phage-displayed combinatorial peptidelibraries. Chem Biol 2000; 7:17–25.

4. Kay BK, Hamilton PT. Identification of enzyme inhibitors fromphage-displayed combinatorial peptide libraries. Comb ChemHigh Throughput Screen 2001; 4:535–543.

5. Scott JK, Huang SF, Gangadhar BP, Samoriski GM, Clapp P,GrossRA,TaussigR,SmrckaAV.Evidence thataprotein–proteininteraction ‘hot spot’ on heterotrimeric G protein betagammasubunits is used for recognition of a subclass of effectors.EMBO J 2001; 20:767–776.


6. Paige LA, Christensen DJ, Gron H, Norris JD, Gottlin EB,Padilla KM, Chang CY, Ballas LM, Hamilton PT, DP McDon-nell, Fowlkes DM. Estrogen receptor (ER) modulators eachinduce distinct conformational changes in ER alpha and ERbeta. Proc Natl Acad Sci USA 1999; 96:3999–4004.

7. Chang C, Norris JD, Gron H, Paige LA, Hamilton PT, KenanDJ, Fowlkes D, McDonnell DP. Dissection of the LXXLLnuclear receptor-coactivator interaction motif using combina-torial peptide libraries: discovery of peptide antagonists ofestrogen receptors alpha and beta. Mol Cell Biol 1999;19:8226–8239.

8. Wrighton NC, Farrell FX, Chang R, Kashyap AK, Barbone FP,Mulcahy LS, Johnson DL, Barrett RW, Jolliffe LK, Dower WJ.Small peptides as potent mimetics of the protein hormoneerythropoietin. Science 1996; 273:458–464.

9. Cwirla SE, Balasubramanian P, Duffin DJ, Wagstrom CR,Gates CM, Singer SC, Davis AM, Tansik RL, Mattheakis LC,Boytos CM, Schatz PJ, Baccanari DP, Wrighton NC, BarrettRW, Dower WJ. Peptide agonist of the thrombopoietin receptoras potent as the natural cytokine. Science 1997; 276:1696–1699.

10. Fairbrother WJ, Christinger HW, Cochran AG, Fuh G, KeenanCJ, Quan C, Shriver SK, Tom JY, Wells JA, Cunningham BC.Novel peptides selected to bind vascular endothelial growthfactor target the receptor-binding site [in process citation].Biochemistry 1998; 37:17754–17764.

11. Han Y, Kodadek T. Peptides selected to bind the Gal80 repres-sor are potent transcriptional activation domains in yeast.J Biol Chem 2000; 275:14979–14984.

12. Gee SH, Sekely SA, Lombardo C, Kurakin A, Froehner SC, KayBK. Cyclic peptides as non-carboxyl-terminal ligands of syn-trophin PDZ domains. J Biol Chem 1998; 273:21980–21987.

13. Fuh G, Pisabarro MT, Li Y, Quan C, Lasky LA, Sidhu SS.Analysis of PDZ domain-ligand interactions using carboxyl-terminal phage display. J Biol Chem 2000; 275:21486–21491.

14. Linn H, Ermekova KS, Rentschler S, Sparks AB, Kay BK,Sudol M. Using molecular repertoires to identify high-affinity

378 Carlson et al.

peptide ligands of the WW domain of human and mouse YAP.Biol Chem 1997; 378:531–537.

15. Sperinde JJ, Choi SJ, Szoka FC Jr. Phage display selection ofa peptide DNase II inhibitor that enhances gene delivery.J Gene Med 2001; 3:101–108.

16. Laskowski RA, Luscombe NM, Swindells MB, Thornton JM.Protein clefts in molecular recognition and function. ProteinSci 1996; 5:2438–2452.

17. Mattos C, Ringe D. Locating and characterizing binding siteson proteins. Nat Biotech 1996; 14:595–599.

18. Ringe D, Mattos C. Analysis of the binding surfaces ofproteins. Med Res Rev 1999; 19:321–331.

19. DeLano WL, Ultsch MH, deVos AM, Wells JA. Convergentsolutions to binding at a protein–protein interface. Science2000; 287:1279–1283.

20. Gron H, Hyde-DeRuyscher R. Peptides as tools in drug discov-ery. Curr Opin Drug Discov Devel 2000; 3:636–645.

21. Christensen DJ, Gottlin EB, Benson RE, Hamilton PT. Phagedisplay for target-based antibacterial drug discovery. DrugDiscov Today 2001; 6:721–727.

22. Takahashi S, Kuzuyama T, Watanabe H, Seto H. A1-deoxy-d-xylulose 5-phosphate reductoisomerase catalyzingthe formation of 2-C-methyl-d-erythritol 4-phosphate in analternative nonmevalonate pathway for terpenoid biosynth-esis. Proc Natl Acad Sci USA 1998; 95:9879–9884.

23. Gottlin EB, Benson RE, Conary S, Antonio B, Duke K, PayneES, Ashraf SS, Christensen DJ. High throughput screen forinhibitors of 1-deoxy-d-xylulose 5-phosphate reductoisomeraseby surrogate ligand competition. J Biomol Screen 2003; 8(3):332–339.

24. Jacobi A, Baum H, Nakano T, Annand RR, Anderson S,Breunig J, Bowes S, Goldenblatt P, Ashraf SS, Gron H,Hamilton PT, Christensen DJ, Loferer H. Isoprenoid bio-synthesis as a novel antibacterial target. Abstracts of the41st Interscience Conference on Antimicrobial Agents andChemotherapy 247 [abstr 2124], 2001.


25. Tao J, Wendler P, Connelly G, Lim A, Zhang J, King M, Li T,Silverman JA, Schimmel PR, Tally FP. Drug target validation:lethal infection blocked by inducible peptide. PNAS USA 2000;97:783–786.

26. Benson RE, Gottlin EB, Christensen DJ, Hamilton PT. Intra-cellular expression of peptide fusions for demonstration ofprotein essentiality in bacteria. Antimicrobiol AgentsChemotherapy 2003; 47(9):2875–2881.

27. Drews J. Drug discovery: a historical perspective. Science2000; 287:1960–1964.

28. Klabunde T, Hessler G. Drug design strategies for targetingG-protein-coupled receptors. Chembiochem 2002; 3:928–944.

29. Wise A, Gearing K, Rees S. Target validation of G-proteincoupled receptors. Drug Discov Today 2002; 7:235–246.

30. Ramer JK, Blaesius R, Fredericks ZL, Roques CN, NicholsonW, Barnett TR, Hamilton PT. Assay of G protein-coupledreceptors using conformation-specific peptide probes that bindG? subunits. Manuscript in preparation.

31. Gnagey A, Ellis J. Allosteric regulation of the binding of[3H]acetylcholine to m2 muscarinic receptors. BiochemPharmacol 1996; 52:1767–1775.

32. Rosenfeld MG, Glass CK. Coregulator codes of transcriptionalregulation by nuclear receptors. J Biol Chem 2001; 276:36865–36868.

33. Coleman KM, Smith CL. Intracellular signaling pathways:nongenomic actions of estrogens and ligand-independent activa-tion of estrogen receptors. Front Biosci 2001; 6:D1379–D1391.

34. Jordan V. The secrets of selective estrogen receptor modula-tion: Cell-specific coregulation. Cancer Cell 2002; 1:215–217.

35. Heery DM, Kalkhoven E, Hoare S, Parker MG. A signaturemotif in transcriptional co-activators mediates binding tonuclear receptors [see comments]. Nature 1997; 387:733–736.

36. McInerney EM, Rose DW, Flynn SE, Westin S, Mullen TM,Krones A, Inostroza J, Torchia J, Nolte RT, Assa-Munt N,Milburn MV, Glass CK, Rosenfeld MG. Determinants of

380 Carlson et al.

coactivator LXXLL motif specificity in nuclear receptortranscriptional activation. Genes Dev 1998; 12:3357–3368.

37. Feng W, Ribeiro RC, Wagner RL, Nguyen H, Apriletti JW,Fletterick RJ, Baxter JD, Kushner PJ, West BL. Hormone-dependent coactivator binding to a hydrophobic cleft onnuclear receptors. Science 1998; 280:1747–1749.

38. Mak H, Hoare S, Henttu P, Parker M. Molecular determinantsof the estrogen receptor–coactivator interface. Mol Cell Biol1999; 19:3895–3903.

39. Brzozowski AM, Pike AC, Dauter Z, Hubbard RE, Bonn T,EngstromO, Ohman L, Greene GL, Gustafsson JA, Carlquist M.Molecular basis of agonism and antagonism in the oestrogenreceptor. Nature 1997; 389:753–758.

40. Shiau AK, Barstad D, Loria PM, Cheng L, Kushner PJ, AgardDA, Greene GL. The structural basis of estrogen receptor=coactivator recognition and the antagonism of this interactionby tamoxifen. Cell 1998; 95:927–937.

41. Bramlett KS, Wu Y, Burris TP. Ligands specify coactivatornuclear receptor (NR) box affinity for estrogen receptorsubtypes. Mol Endocrinol 2001; 15:909–922.

42. WongCW,KommB, Cheskis BJ. Structure–function evaluationof ER alpha and beta interplay with SRC family coactivators.ER selective ligands. Biochemistry 2001; 40:6756–6765.

43. Kraichely DM, Sun J, Katzenellenbogen JA, KatzenellenbogenBS. Conformational changes and coactivator recruitment bynovel ligands for estrogen receptor-alpha and estrogen recep-tor-beta: correlations with biological character and distinctdifferences among SRC coactivator family members.Endocrinology 2000; 141:3534–3545.

44. Ko L, Cardona GR, Iwasaki T, Bramlett KS, Burris TP, ChinWW. Ser-884 adjacent to the LXXLL motif of coactivator TRBPdefines selectivity for ERs and TRs. Mol Endocrinol 2002;16:128–140.

45. Xu HE, Stanley TB, Montana VG, Lambert MH, Shearer BG,Cobb JE, McKee DD, Galardi CM, Plunket KD, Nolte RT,Parks DJ, Moore JT, Kliewer SA, Willson TM, Stimmel JB.Structural basis for antagonist-mediated recruitment of


nuclear co-repressors by PPARalpha. Nature 2002; 415:813–817.

46. Jung D, Lee S, Lee J. Agonist-dependent repression mediatedby mutant estrogen receptor alpha that lacks the activationfunction 2 core domain. J Biol Chem 2001; 276:37280–37283.

47. He B, Minges JT, Lee LW, Wilson EM. The FXXLF motifmediates androgen receptor-specific interactions with cor-egulators. J Biol Chem 2002; 277:10226–10235.

48. Li D, Wang F, Samuels H. Domain structure of the NRIF3family of coregulators suggests potential dual roles intranscriptional regulation. Mol Cell Biol 2001; 21:8371–8384.

49. Warnmark A, Treuter E, Gustafsson J, Hubbard R, Brzo-zowski A, Pike A. Interaction of transcriptional intermediaryfactor 2 nuclear receptor box peptides with the coactivatorbinding site of estrogen receptor alpha. J Biol Chem 2002;277:21862–81868.

50. Norris JD, Paige LA, Christensen DJ, Chang CY, HuacaniMR, Fan D, Hamilton PT, Fowlkes DM, McDonnell DP. Pep-tide antagonists of the human estrogen receptor. Science1999; 285:744–746.

51. Finkel T, Duc J, Fearon E, Dang C, Tomaselli G. Detection andmodulation in vivo of helix-loop-helix protein–protein interac-tion. J Biol Chem 1993; 268:5–8.

52. Hirsimaki P, Aaltonen A, Mantyla E. Toxicity of antiestrogens.Breast J 2002; 8:92–96.

53. Witcher M, Yang X, Pater A, Tang SC. BAG-1 p50 isoforminteracts with the vitamin D receptor and its cellular overex-pression inhibits the vitamin D pathway. Exp Cell Res 2001;265:167–173.

54. Boruk M, Savory JG, Hache RJ. AF-2-dependent potentiationof CCAAT enhancer binding protein beta-mediated transcrip-tional activation by glucocorticoid receptor. Mol Endocrinol1998; 12:1749–1763.

55. Jung DJ, Na SY, Na DS, Lee JW. Molecular cloning andcharacterization of CAPER, a novel coactivator of activatingprotein-1 and estrogen receptors. J Biol Chem 2002;277:1229–1234.

382 Carlson et al.

56. Lu ML, Schneider MC, Zheng Y, Zhang X, Richie JP. Caveolin-1 interacts with androgen receptor. A positive modulator ofandrogen receptor mediated transactivation. J Biol Chem2001; 276:13442–13451.

57. Rao J, Lee P, Benzeno S, Cardozo C, Albertus J, Robins DM,Caplan AJ. Functional interaction of human Cdc37 with theandrogen receptor but not with the glucocorticoid receptor.J Biol Chem 2001; 276:5814–5820.

58. Sauve F, McBroom LD, Gallant J, Moraitis AN, Labrie F,Giguere V. CIA, a novel estrogen receptor coactivator with abifunctional nuclear receptor interacting determinant. MolCell Biol 2001; 21:343–353.

59. Hulkko SM, Wakui H, Zilliacus J. The pro-apoptotic proteindeath-associated protein 3 (DAP3) interacts with the glucocor-ticoid receptor and affects the receptor function. Biochem J2000; 349(Pt 3):885–893.

60. Chen J,Kaul S, Simons SS Jr. Structure=activity elements of themultifunctional protein, GMEB-1. characterization of domainsrelevant for the modulation of glucocorticoid receptor transacti-vation properties. J Biol Chem 2002; 277:22053–22062.

61. Kaul S, Blackford JA Jr, Chen J, Ogryzko VV, Simons SS Jr.Properties of the glucocorticoid modulatory element bindingproteins GMEB-1 and -2: potential new modifiers of gluco-corticoid receptor transactivation and members of the familyof KDWK proteins. Mol Endocrinol 2000; 14:1010–1027.

62. SharmaM, Zarnegar M, Li X, Lim B, Sun Z. Androgen receptorinteracts with a novel MYST protein, HBO1. J Biol Chem 2000;275:35200–35208.

63. Saji S, Okumura N, Eguchi H, Nakashima S, Suzuki A, Toi M,Nozawa Y, Hayashi S. MDM2 enhances the function ofestrogen receptor alpha in human breast cancer cells. BiochemBiophys Res Commun 2001; 281:259–265.

64. Schulz M, Schneider S, Lottspeich F, Renkawitz R, Eggert M.Identification of nucleolin as a glucocorticoid receptorinteracting protein. Biochem Biophys Res Commun 2001;280:476–480.


65. Lee SR, Ramos SM, Ko A, Masiello D, Swanson KD, Lu ML,Balk SP. AR and ER interaction with a p21-activated kinase(PAK6). Mol Endocrinol 2002; 16:85–99.

66. Wang Y, Yin L, Hillgartner FB. The homeodomain proteinsPBX and MEIS1 are accessory factors that enhance thyroidhormone regulation of the malic enzyme gene in hepatocytes.J Biol Chem 2001; 276:23838–23848.

67. Zhou D, Chen S. PNRC2 is a 16kDa coactivator that interactswith nuclear receptors through an SH3-binding motif. NucleicAcids Res 2001; 29:3939–3948.

68. Delage-Mourroux R, Martini PG, Choi I, Kraichely DM,Hoeksema J, Katzenellenbogen BS. Analysis of estrogen recep-tor interaction with a repressor of estrogen receptor activity(REA) and the regulation of estrogen receptor transcriptionalactivity by REA. J Biol Chem 2000; 275:35848–35856.

69. Chipuk JE, Cornelius SC, Pultz NJ, Jorgensen JS, BonhamMJ, Kim SJ, Danielpour D. The androgen receptor repressestransforming growth factor-beta signaling through interactionwith Smad3. J Biol Chem 2002; 277:1240–1248.

70. Stoner M, Wang F, Wormke M, Nguyen T, Samudio I, VyhlidalC, Marme D, Finkenzeller G, Safe S. Inhibition of vascularendothelial growth factor expression in HEC1A endometrialcancer cells through interactions of estrogen receptor alphaand Sp3 proteins. J Biol Chem 2000; 275:22769–22779.

71. Watanabe M, Yanagisawa J, Kitagawa H, Takeyama K,Ogawa S, Arao Y, Suzawa M, Kobayashi Y, Yano T, YoshikawaH, Masuhiro Y, Kato S. A subfamily of RNA-binding DEAD-box proteins acts as an estrogen receptor alpha coactivatorthrough the N-terminal activation domain (AF-1) with anRNA coactivator, SRA. EMBO J 2001; 20:1341–1352.

72. Makino Y, Yoshikawa N, Okamoto K, Hirota K, Yodoi J,Makino I, Tanaka H. Direct association with thioredoxinallows redox regulation of glucocorticoid receptor function. JBiol Chem 1999; 274:3182–3188.

384 Carlson et al.

10

Engineering Protein Foldingand Stability

MIHRIBAN TUNA and DEREK N. WOOLFSON

Department of Biochemistry, School of LifeSciences, University of Sussex,

Falmer, U.K.

I. PROTEIN REDESIGN AND DESIGN

Protein design tests our understanding of how protein sequ-ence relates to structure and function; that is, our progresstowards addressing the informational aspect of the protein-folding problem. In addition, successful protein-design exer-cises provide tools, concepts, and rules for: (1) engineeringexisting protein scaffolds (an area that we refer to as proteinengineering or protein redesign), and (2) creating altogethernew protein structures and functions (de novo protein design).This second endeavor is the ultimate quest for some proteindesigners, but it is difficult and, by and large, can only be

385

achieved for a limited number of relatively straightforwardprotein-folding motifs—for instance, coiled coils and zinc-finger peptides—for which good sequence-to-structure rela-tionship are available (1); although, very recently, Bakerand colleagues (2) describe the successful computer-aideddesign of a completely novel globular protein.

In general, however, de novo protein design attempts forsuch targets tend to deliver partly folded, molten-globuleensembles rather than specific, unique structures, thoughthe secondary structure content and even chain topologymay be correct. Thus, for globular structures alternativeapproaches are required in design: one extreme is to choosean appropriate stably folded scaffold as a starting point andengineer changes in a rational and usually stepwise mannertoward a target with altered and=or improved properties. Atthe other extreme, the starting protein, which may be foldedor not, is subjected to combinatorial mutagenesis to createlibraries of related protein sequences from which folded,stable, and perhaps even functional proteins are selected.The mutagenesis employed can either be random across partor all of the sequence, or saturation mutagenesis targeted atspecific residues chosen by the researcher on some rationalbasis.

Choosing where and how to mutate is not always clearand herein lies a major problem in redesigning and designingglobular proteins. In part, the problem is that, from an experi-mental point of view, for a given protein, it is not possible tocreate redundant libraries in which more than about eightresidues are mutated to all possible amino acids. Thisproblem has been solved to some extent by computationalmethods in which very large numbers of mutants can beassessed against a targeted structural model very rapidly(3–5). A difficulty with this in silico approach, however, isthat, compared with combinatorial experimental methods,only a relatively small number of the generated sequencescan be made and tested experimentally. In addition, foldedproteins are highly cooperative units; that is, their structuresand stabilities are determined by many interdependent (i.e.,cooperative) noncovalent interactions. In this respect, it is

386 Tuna and Woolfson

hard to fully disentangle the main contributors to proteinstability. Thus, any multiple mutagensis experiment carriesthe risk that as much damage can be done as good; in otherwords, many of the mutants in any given library will fail tofold altogether.

As said above, it is widely accepted that the acquisition of awell-packed and complementary hydrophobic core is a generalrequirement for specifying the fold of a globular protein and formaintaining its stability (6). Indeed, the lack of stability ofmany early protein designs, particularly those for four-helix-bundle proteins, has been put down to a lack of specificity inthe prescribed hydrophobic cores (7–9). Thus, tackling theprotein design problem from the inside out may seem like areasonable tack. Indeed, over the past decade, both computer-based and experimental protein-design efforts have focusedondevelopingmethods for optimizing hydrophobic core packing(3–5). This in silico work has been very successful. However,this chapter focuses on experimental efforts in this area. Thetricks here are: (1) to target a restricted (rationally chosen)set of sites for mutagenesis; and (2) to design selection methodsthat allow the small fraction of competently folded and stablestructures from the vast majority of mutants that, even withrational targeting, may fail to fold. We begin by summarizingthe main experimental efforts that preceded phage-displaywork in this area. However, as we shall also describe, othershave had considerable success in engineering protein stabilityby leaving the hydrophobic core well alone and optimizingsurface residues; presumably in these cases a higher proportionof the mutant proteins do fold.

II. EARLY COMBINATORIAL STUDIES AIMEDAT REPACKING THE CORES OF PROTEINS

II.A. Repacking the Cores of Natural Proteins

The studies of Lim et al. (10–12) on the N-terminal domainof l-repressor provide a landmark in this area. A library ofl-repressor mutants is created by saturation mutagenesis toNNfC,Gg codons (i.e., to give all of the possible 20 proteino-genic amino acids) at seven hydrophobic core positions. An

Engineering Protein Folding and Stability 387

in vivo functional assay is used to screen for competentlyfolded mutants of the domain: the DNA for the library istransformed into Escherichia coli, which are subsequentlychallenged with a strain of phage l lacking a functionall-repressor gene. Thus, only those clones that expressfunctional l-repressor survive. The assertion is that in orderto function the mutant proteins must be folded correctly.The studies conclude that the core of this particular proteindomain is tolerant to substitutions and as such is plastic.The main constraint on achieving functional mutants is themaintenance of hydrophobic residues within the core and,albeit to a lesser extent, the preservation of steric andcore-volume constraints.

Axe et al. (13) repack the hydrophobic core of the ribonu-clease barnase to investigate whether unique native-likepacking is essential for maintenance of protein function. Inthis case, the library is made by mutagenesis of 12 of the 13hydrophobic core sites to combinations of Phe, Ile, Leu, Met,and Val (FILMV), which are readily generated by the degen-erate codon fA,G,TgTfC,Gg. Again a functional screen is usedto detect functional and presumably folded proteins. Barnaseis extremely toxic to bacteria and the assay developed is extre-mely sensitive as mutants with as low as 0.2% of wild-type(WT) activity kill the host and test positive. Thus, a suppres-sible stop codon within the coding sequence for the barnasemutants is used in these studies, with active variants beingidentified by switching between nonsuppressor and suppres-sor strains of E. coli. As high as 23% of the hydrophobiccore variants maintain enzymatic activity in vivo. Thus,the authors argue that, like the N-terminal domain of l-repressor, the basic function of barnase is very tolerant ofchanges in the hydrophobic core. Hence, in terms of designingor maintaining crude function in proteins, hydrophobicityalone could be sufficient to generate a functional hydrophobiccore; and after this step further mutagenesis and optimiza-tion of the packing could be employed to achieve a better(i.e., unique and more stable) scaffold (13).

One assumption in the studies by Lim and Sauer and byAxe and colleagues is that correct folding is a prerequisite for


correct function. Given the accepted link between proteinsequence, structure, and function established by Anfinsen,this seems reasonable. However, imagine a case where a pro-tein is either unfolded or only partly folded in isolation, butfolds upon binding to its substrate. Such onsite assembly iswell established in the area of protein–nucleic acid interac-tion. Furthermore, this idea of disorder-to-order transitionsassociated with protein function is gathering steam in thearea of natively unfolded proteins (14). In the approachesof Axe et al. and Lim and Sauer, such transitions, if theyoccurred, would give false positives in the sense that certainproteins would not necessarily have to fold independently inorder to elicit a functional response.

II.B. Creating Novel Proteins Using BinaryPatterns of Hydrophobic and PolarAmino Acids

Hecht and coworkers have championed a very differentapproach in which focused libraries for de novo designed pro-teins are created based on straightforward binary, or HP pat-terns of hydrophobic (H) and polar (P) residues. Put simply,patterns of the type HPHPHP or HPPHPPP separatedby turn-promoting sequences are used to guide folding tob-strand and a-helical elements of secondary structure,respectively. Limited amino-acid repertoires are also used,specifically P ¼ K, H, Q, N, E, or D and H ¼ F, I, L, M, or V.Hecht and his colleagues have used this approach to generatefour-helix-bundle motifs and, more recently, six-stranded b-sheet models. The approach differs considerably from thosethat form the main focus of this chapter because the librariesare ‘‘not subjected to high-throughput screens or directed evo-lution’’; although one imagines that some selection is at playwithin E. coli, the host into which the DNA libraries aretransformed and expressed. Rather, clones are selected purelyon the basis of expression of the mutant proteins. Severaliterations of the four-helix-bundle designs have beenperformed and generate proteins that are water soluble, havea-helical structure and some show native-like characteristics


(15–19). Nonetheless, in the earlier studies at least, mostexhibited characteristics typical of molten-globule structures.However, this work has recently culminated with the determi-nation of an NMR solution structure for a four-helix-bundleprotein retrieved from a binary library (20), and the observa-tion that some of these proteins exhibit enzyme-like activity(21).

Interestingly, this approach does not appear to extendreadily to the generation of globular b-structured scaffolds.Using libraries created to generate amphipathic b-strandsseparated by turn regions, the group retrieved proteins thatformed amyloid-like fibrous assemblies (22). Encouragingly,however, Wang and Hecht (23) have successfully managedto prevent amyloid-like fibrillogenesis in one of the selectedb-structured mutants by incorporating rules proposed byRichardson and Richardson (24) for capping the edges ofb-sheets.

To finish this section, we should like to mention the workof Harbury and colleagues who describe the combinatorialdissection of the TIM-barrel structure, which is adopted byapproximately 10% of all enzymes. This structure effectivelyhas two hydrophobic cores, which Harbury and coworkers(25) find to respond differently to mutation. The outer corebetween the a-helices and the b-barrel tolerates mutations,whereas the inner core, which is more conserved, is moresensitive to changes. This second finding is particularlyimportant as it contrasts with most of the foregoing work inthis area and that described above, which suggests thathydrophobic cores are generally plastic and permissive of(hydrophobic) mutations. This point, which might be termed‘‘oil-drop vs. jigsaw-puzzle models’’ for the packing of hydro-phobic cores, is a theme that we will return to later whenwe describe studies on ubiquitin (26,27).

III. PHAGE DISPLAY IN ENGINEERINGPROTEIN STABILITY

Phage display has been proved an effective tool in thearmoury of the protein engineer. In this method, a protein


of interest is displayed on the surface of (usually) filamentousbacteriophage via fusion to one of the viral coat proteins (seeChapter 2). This carries a number of advantages. First, theprotein is linked to the DNA that encodes it, which is encasedwithin the virus particle. Thus, phenotype and genotype arephysically linked, allowing the displayed proteins to be iden-tified through DNA sequencing as necessary. Second, usingstandard mutagenesis techniques, the gene for the displayedprotein can be mutated to create a library of variants. Each ofthese variants can then be displayed on distinct phage parti-cles (to give what are termed ‘‘protein-phage’’) and are, hence,linked to their own encoding DNA. Third, using suitableselection methods, protein-phage variants with specificallysought properties can be retrieved from the pool. These canbe amplified by infection into E. coli. At this point, the phagecan be recovered for further rounds of selection and amplifica-tion, or clones can be identified by DNA sequencing of thephagemid DNA within the E. coli. All in all, the phage-displayapproach offers the opportunity of exploring a large numberof sequences simultaneously. This method has most widelybeen applied to select protein variants, particularly thosefor antibody fragments, based on their ability to bind specifictargets.

III.A. Non-Protease-Based Applications

Baker and colleagues (28) use binding to select stably foldedproteins from a phage-displayed library of the IgG-bindingdomain of peptostreptococcal protein L. The selection ratio-nale is that the correct folding of the domain is required forIgG binding. The group shows that, even without disruptionor alteration to the binding interface, if the structure andfolding of the domain is compromised IgG binding isabolished. In this case, this clearly argues against the afore-mentioned potential caveat about cooperative folding andbinding (i.e., onsite folding, or disorder-to-order transitions).The group displays the domain via the major coat protein,g8p, and mutate 14 residues of the N-terminal b-hairpin ofthe protein. As noted by the authors, the mutagenesis and


selections presented are too limited to allow conclusions aboutprotein folding to be made. Thus, the experiments should beconsidered as proof of concept aimed at developing a methodto study the sequence determinants of protein folding forstructures that have a binding site that can be exploited inselection.

Chakravarty et al. (29) have taken a slightly differentapproach to achieve protein stabilization. They use fragmentcomplementation in combination with phage display toimprove the thermostability of RNase S. Specifically, thelarge proteolytic fragment S-protein (residues 21–124) is usedto select tight binders from a random 15-mer peptide librarypresented on phage. One selected peptide complements S-protein to give a complex with melting temperature 10�Chigher than the WT.

III.B. Combining Phage Display and Proteolysisin Protein Engineering and Design

In this chapter, we are primarily interested in selection meth-ods in which the key selection step does not involve the abilityof displayed protein itself to bind some prescribed target.Therefore, the question is what alternative selection strate-gies can be conceived to retrieve proteins on the basis of fold-ing and stability alone, rather than based on some functionsuch as binding? Such a method would immediately removeany concern over false positives resulting from onsite assem-bly of the selected proteins; it would provide a route for asses-sing protein–sequence relationships in proteins, and presentnew possibilities for designing new proteins from scratch freeof constraints imposed by function. One option, which hasconsidered independently by at least three groups (30–32),is that proteolysis could be combined somehow with phage-display selection to remove proteins that are either unfoldedor only partly folded.

In many respects, the seminal publication fromMatthews and Wells (33) provides the forerunner for such astrategy. These authors describe the concept of substratephage in which a library of peptides is tethered between an


affinity tag and the gene-3 minor coat protein (g3p). Theresulting phage are bound to a surface and treated with pro-tease. As phage are themselves protease insensitive, choosinga similarly resistant affinity tag means that this methodreports directly on the protease sensitivity of the peptide lin-ker: i.e., phage released quickly from the surface harboredgood protease substrates, whereas those that remain boundharbored poor substrates. In this way, Matthews and Wellsare able to identify pools of sensitive and resistant peptidesequences for a variant of subtilisin and factor X(a). In thismethod and those that followed (described below), the bindingproperties of the affinity tag are of secondary importance andthe selection is predicated by the cleavage (or not) of the tagfrom the phage, which depends only on the susceptibility (orotherwise) of the peptide linker itself.

Initially, three independent groups reported successfulstudies utilizing phage display and protease selection forprobing and=or increasing the stability of proteins (30–32).In addition, these groups and a number of others have appliedthe approach in protein engineering and design (34,35); theselater studies will be described in Sec. V of this chapter. All ofthe selection methods are based purely on maintaining thestructural integrity of the target proteins, and there is norequirement for any other assayable function. In each case,the selection rationale is that stable, fully folded proteinsare more resistant to attack by proteases than partially foldedor unfolded proteins (36).

In 1998, Kristensen and Winter (30) proposed that pro-teolysis could be used to select stably folded proteins fromphage-displayed libraries. This method is a variant of selec-tively infective phage, SIP: a peptide or protein of interest iscloned between the D2 and D3 domains of g3p. D3 anchorsg3p to the viral surface, while the D1 and D2 domainsare required for infection of E. coli. Thus, cleavage of theD2–D3 linker renders the phage noninfective, which meansthat DNA carried by such phage is not propagated in thephage-display and selection experiment (Fig. 1). The groupdemonstrate that protein folding and protease sensitivityare linked by monitoring and comparing phage infectivity of


various peptide-phage and protein-phage fusions—specifically,a protease-sensitive peptide, two variants of barnase with dif-ferent stabilities and the villin headpiece—after treatment withtrypsin over a range of temperature. In addition, they showthat this provides the basis for discriminating peptides andproteins of different stabilities (i.e., selection) by tracking

Figure 1 (a) Selectively infective phage takes advantage of thethree-domain structure of the minor coat protein (g3p) of phage.The C-terminal domain anchors the protein in the viral coat,whereas the N-terminal domains are responsible for binding andinfection into E. coli. Cloning a library into the flexible linker beforethe C-terminal domains allows protease-based selection becauseproteolysis of the insert removes the N-terminal domains and pre-vents infection into E. coli. This selects against unstable inserts(30,31). (b) Alternatively, an uninterrupted g3p can be used as follows:His-tag–target–g3p–phage. This allows intact protein-phagefusions to be tethered to Ni-coated surfaces, which can be washedwith protease to remove phage harboring unstable linkers (32). Inthis case, selection can be monitored directly by SPR in Biacore,which allows many conditions to be tested quickly and individualclones to be compared. Alternatively, Ni-NTA-agarose beads canbe used for large-scale selections. (Reproduced from Ref. 1.)


phage retrieved after protease challenges on mixtures of: (1)the two barnase variants and (2) the protease-sensitivepeptide and the villin headpiece. This paper is also notablein a number of other respects: first, the authors demonstratethe ability of filamentous phage to endure a variety of insultsincluding incubations over a range of pH (pH 2.2–12, 37�C,30min) and in the presence of urea (60�C, 60min) orguanidine hydrochloride (37�C, 90min); appreciable loss inphage infectivity is only reported for experiments using 5Mguanidine hydrochloride and higher. Similarly, resistanceto a range of proteases is also reported. In this case, phageinfectivity is only reduced after treatment with subtilisin,which is known to cleave g3p. In addition, to assist the aboveexperiments, the authors describe a protease-cleavablehelper phage (KM13) and a phagemid vector system. Theseare created by introducing a protease-cleavable linkerbetween the D2 and D3 domains of g3p. The KM13 helperphage is used to rescue target protein-phage, which meansthat after protease selection only those phage with intactD1–D2–protein–D3 fusions will be infective and, hence, willbe selected.

In the same year, Sieber et al. (31) describe a similarSIP-based approach for selecting stably folded proteins fromphage-displayed libraries using protease selection. The groupdubs their method Proside for ‘‘protein stability increased bydirected evolution.’’ Like Kristensen and Winter, the groupdescribes a number of important and useful control andexploratory experiments. Specifically, they also demonstratethe stability of fd phage over a range of solution conditions,which may be of use in stability based selections of protein-phage; they show that protease selection varies with differentproteases, namely trypsin, chymotrypsin, pepsin, or protei-nase K, with trypsin showing the lowest activity and protei-nase K the highest; and they show that by fine-tuning theconditions of the selection process variants with marginallydifferent stabilities (differing by �1 kJ=mol) can be distin-guished. Finally, the group demonstrates selection from amodest library of variants of RNaseT1(4A) with mutationsof surface residues. RNaseT1(4A) is a destabilized variant of


RNaseT1 in which the four Cys residues are mutated to Alato remove the two disulfide bonds. Positions, Ser17, Asp29,and Tyr42 of the variant are targeted by saturation mutagen-esis. After four rounds of protease selection the preferred resi-dues emerge at these sites, namely Ala, Leu, and Phe,respectively. Impressively, when the identified mutationsare transferred to WT RNaseT1, with its disulfide bondsintact, the resulting proteins are also stabilized significantly(by almost 10�C increases in their TMs). This is an importantresult, as it shows that surface engineering of proteins can beused to improve stability of proteins. Furthermore, it demon-strates that mutations identified in this way can be trans-ferred between backgrounds, which is unlikely to be true ofconstellations of stabilizing residues discovered through core-directed design where residue–residue contacts are greater.

We published our own work in 1999, and we combinedphage display and protease selection in a different way (Figs.1 and 2) (32). In this case, the target protein is displayed onbacteriophage more traditionally, i.e., as an N-terminal fusionto the minor coat protein g3p. In addition, a histidine tag(either hexa- or deca-His) is incorporated at the N-terminusof target to provide an affinity tag for metal-based captureof the protein-phage onto surfaces. After the protein-phageare immobilized onto nickel-coated surfaces, they are chal-lenged with a protease or protease mixtures. This results incleavage of any unstable protein linkers and allows phageharboring such linkers to be washed away from the support.The remaining protein-phage are then eluted from the sup-port using imidazole or EDTA washes, or with a change ofpH of the buffer. The freed phage are then amplified forfurther rounds of selection and=or identified by DNA sequen-cing. We find that a particular advantage of this approach isthat the proteolysis reaction can be monitored in real timeby following the release of phage by surface plasmonresonance (SPR) in Biacore. Note that this type of selectioncan also be done in solution with intact (protease-resistantprotein-phage) being pulled down after proteolysis. As detailedin the next section, we have applied this method in anapproach that we refer to as core-directed protein design (1)


to repack the hydrophobic core of ubiquitin (27). In this case,we find that, even though we mutated one half of theprotein’s hydrophobic core simultaneously, all eight of thetargeted residues showed a strong preference for WT orWT-like residues in the selected library.

IV. A WORKED EXAMPLE: REPACKINGTHE HYDROPHOBIC CORE OF UBIQUITIN

We have combined the ideas of core-directed protein designand protease-based phage-display selection to repack thehydrophobic core of ubiquitin (27). To achieve this, the genefor mammalian ubiquitin was cloned between those for an(N-terminal) hexahistidine tag and gene III in a phage-display vector, pCANTAB B2. This construct included two

Figure 2 Details of the protease-selection method developed inthe Woolfson laboratory (32). (1) Protein-phage are captured ontonickel-derivatized support via N-terminal histidine tags. (2) Boundprotein-phage are challenged with protease, phage-displaying pro-teins that are proteolyzed are washed away from the support. (3)Protein-phage that resist proteolysis are eluted from the nickelsupport and (4) amplified. (Reproduced from Ref. 49.)


suppressible stop codons prior to gene III to allow expressionof the hexahistidine-tagged protein both as a fusion to g3pand free of bacteriophage by using suppressor and nonsup-pressor strains of E. coli, respectively. Residues corres-ponding to approximately half of the hydrophobic core ofubiquitin were mutated to combinations of hydrophobic sidechains to produce a phage library. Specifically, residues 1, 3,5, 15, 17, 26, and 30 were targeted for mutagenesis to combi-nations of FILMV using the aforementioned DTS degeneratecodon (fA,G,TgTfC,Gg). To avoid a bias of WT ubiquitin—which is exceptionally stable with respect to unfolding andproteolysis—a nobbled template was used for the combinator-ial mutagenesis; namely, Met 1 was replaced by Ala and theremaining targeted sites were replaced by Leu to give an‘‘AL7’’ mutant. Theoretically at least, this combination ofAla and Leu has a similar volume to the WT targeted coreresidues. This mutant was destabilized with respect toproteolysis. The AL7 plasmid was used as the template forsticky-feet mutagenesis (37) to create the library of ubiquitinhydrophobic core mutants. This library potentially contained1.17� 107 different DNAs and 390,625 protein variantsincluding the WT sequence. Experimentally, DNA sequen-cing revealed that the naıve library was 48% AL7 parentand contained approximately 7�106, corresponding toapproximately a 20-fold redundancy of the potential proteinmutants.

Turning to the phage display and selection, two differentnickel-coated surfaces were tested for immobilizing theprotein-phage and performing protease selection on thelibrary of ubiquitin hydrophobic core mutants: nickel-NTAsensor chips for SPR (Biacore) and nickel-NTA agarose beads(QIAgen).

IV.A. Following Protease Selection by SurfacePlasmon Resonance

Surface plasmon resonance (SPR) technology has been devel-oped specifically for monitoring biomolecular interactions(38). The basic principle of this method is that changes in mass


at a surface result in a change in refractive index of thesolution above the surface. This gives rise to an SPR effectat the surface, which can be measured by monitoring changesin a laser beam reflected by the surface. The SPR response,measured in resonance units (RU), is proportional to theamount of material bound to the surface. In a typical SPRexperiment, one of the interactants is immobilized to the sur-face of a sensor chip in a microflow cell into which the otherinteractant(s) is injected. Any interactions lead to a changein the effective mass at the surface and hence a refractiveindex change. Typically, the response is followed with timeto give a record, sensogram, of the interactions on the sensorchip. The method has been used to study interactions betweenbiomolecules such as peptides, nucleic acids, carbohydrates,lipids, and small molecules. It is performed in real time anddoes not require labelling. It has mainly been used to evaluatebinding, binding specificity, and kinetics (39), although otherapplications include ligand fishing, epitope mapping, molecu-lar assembly, following purification and small-moleculescreening (40). Various sensor chips have been designed forspecific interactions with ligands. We used sensor chips withnitrilotriacetic acid (NTA) immobilized on a dextran matrixto capture nickel and thence histidine-tagged protein-phage.

We used SPR to follow and to visualize binding and pro-tease cleavage of both monoclonal and library protein-phage(32). This established: (1) specific binding of protein-phageto NTA sensor chips in Biacore; (2) that the binding capacityof such chips was limited, but still useful; and (3) that themethod could be used to follow protease cleavage.

Figure 3 shows a typical sensogram for an SPR experi-ment performed using a Biacore experiment. Both the initialloading of nickel ions and the subsequent binding and releaseof protein-phage are detected in real time using this techni-que. In each case, binding is measured as net change in reso-nance response, in RU, before and after the addition of thenickel or the analyte.

Although we detected some background (nonspecific)binding, we only observed significant binding of protein-phage when both nickel had been loaded onto the chip and


Figure 3 Sensorgram trace for an experiment of the type shownin Fig. 2 and followed by SPR as carried out in a Biacore 2000instrument. At time t1, the resonance intensity (measured in RU)is at a minimum, and the intensity can be normalized to 0, asshown. While buffer containing additional components, such asnickel ions, phage, or enzyme, is passed over the chip, the resonanceintensity will change because of the refractive index change, asshown by the immediate rise in intensity in the figure on addingthe nickel solution. When the running buffer wash is resumed theresonance intensity drops again, with the new equilibrium value(t2) representing the material that bound in the previous step. Oncethe nickel has been bound, the phage are added, which then bindthrough their hexahistidine tags. After the wash is resumed, theintensity increases to t3 indicating bound phage [bound phage�(t3 � t2)]. A further brief increase in resonance intensity indicatesthe change due to the chymotrypsin-containing solution, but this isfollowed by a decline as phage are stripped from the chip by proteo-lysis. After the resumption of washing the new equilibrium value t4is lower than t3, indicating the loss of phage from the surface(cleaved phage �(t3 � t4)). Finally, an EDTA solution strips allnickel ions from the surface, and also any remaining phage and hex-ahistidine tags bound to the ions, restoring the resonance intensity(t5) to the initial value (t1). (Adapted from Ref. 32.)


the protein-phage carried a histidine tag. On this point, wefound that decahistidine tags performed better and morereliably than traditional hexahistidine tags.

The binding capacity of a flowcell on an NTA chipwas determined by injecting varying amount of protein-phage until saturation was reached—i.e., until the SPR signallevelled off. The maximum response observed upon binding toone flowcell on the NTA chip was approximately 250–300 RU.According to the manufacturer (Biacore AB), 1000 RU corre-sponds to �1 ng=mm2 bound material on the sensor chipsurface and the area used for binding of one flowcell on thechip is 1 mm2. Thus, 250 RU corresponds to approximately2.5� 10�10 g of bound protein-phage, which, with the molecu-lar weight of filamentous phage estimated at 1.5� 106 g=mol,gives the maximum number of protein-phage binding withinone flowcell as approximately 107. We also determined thisvalue experimentally by recovering and quantifying thebound phage by infecting log-phase E. coli cells. This experi-ment gave an estimate of 9.6� 106 bound protein-phage ingood agreement with the above calculation.

We used two protein-phage constructs and chymotrypsintreatment to test the utility and specificity of our proposedapproach to selection (32). First, we chose a stable construct,His6-WT-UBQ-phage, harboring WT ubiquitin; although theprotein contains both Tyr and Phe, which are targeted by chy-motrypsin, as well as Leu, which is cleaved to a lesser extent,it resists proteolysis for some time. The tyrosine residuewas included in the mutagenesis to create the ubiquitinlibrary described below. In addition, a negative-control con-struct, His6-FLEXI-phage, was created, which harbored aGly-Ser-based flexible linker peptide with a single Tyr clea-vage site. Figure 4 shows a comparison of the response ofthese two constructs to chymotrypsin treatment. These datawere collected simultaneously from two lanes on a Biacoresensor chip. The dramatic difference in loss of protein-phagefrom the surface between FLEXI and WT-UBQ confirmedthat proteolysis was sensitive to the sequence of the displayedprotein. Similarly, when His6-AL7-UBQ-phage were chal-lenged with chymotrypsin the degree of proteolysis fell in


between that for the FLEXI and WT-UBQ constructs. Weconfirmed and reproduced these results using both continu-ous and pulsed treatments with protease (Figs. 4 and 5).Thus, the protease-selection method discriminated betweenfolded proteins, destabilized protein, and flexible polypeptidechains.

Finally, we followed protease selection of His6-LIBRARY-UBQ-phage by SPR in Biacore; LIBRARY-UBQ was the afore-mentioned library of ubiquitin hydrophobic core mutants.Interestingly, the rate of proteolysis of LIBRARY-UBQ-phagewas initially fast, but then slowed to a rate similar to that forHis6-WT-UBQ-phage. This suggests that the library contains

Figure 4 Sequence dependence of the protease cleavage profile asdetermined by SPR. The solid trace shows the cleavage of phage dis-playing WT ubiquitin, whereas the dashed line is for FLEXI-phage.The data have been normalized such that 100 on the Y-axis isequivalent to 100% of the phage bound immediately before proteoly-sis. After 300 sec of exposure to chymotrypsin, 83% of the WT-UBQ-phage remained bound to the surface of the Biacore chip,in contrast, only 23% of the FLEXI-phage remained. The plainsolid line shows an underivatized control channel treated withchymotrypsin. (Adapted from Ref. 32.)


some poorly folded and protease-susceptible mutants that arecleaved and removed rapidly, and some more resilientmutants that we cleaved slowly like WT ubiquitin.

In summary to this section, we were able to show thatSPR could be used to follow the cleavage and release ofprotein-phage from the surfaces of Biacore sensor chips. Inaddition, we demonstrated that binding to the chips requiredN-terminally His-tagged protein-phage, and that the rate anddegree of proteolysis was related to the integrity of the foldedstructure of the displayed protein. Whilst SPR provides astraightforward method for following protease treatment ofmonoclonal protein-phage, it is clearly limited for performingselection studies from phage-displayed libraries because thebinding capacity of Biacore N-NTA sensor chips is limited to

Figure 5 Pulsed proteolysis of various protein-phage bound to aBiacore sensor chip. In this experiment, changes in SPR signal weremeasured after pulses with 10mm chymotrypsin. After each pulse,washing was resumed to reestablish equilibrium and determinethe amount of phage remaining. The figure shows that cumulativeloss of phage from the surface is plotted against the cumulativeexposure time to protease for WT ubiquitin, library, and the control,FLEXI-phage. (Adapted from Ref. 32.)


approximately 107 protein-phage. Realistically, this limitslibrary sizes to about 106 protein mutants. In addition, inour experience, it is not straightforward to recover reproduci-bly protein-phage from Biacore sensor chips. With thesedrawbacks in mind, we turned to the more traditional Ni-NTA agarose beads (QIAgen) as a support for sequesteringprotein-phage prior to protease selection.

IV.B. Preparative Protease Selection fromPhage-Displayed Libraries

Firstly, we tested that Ni-NTA agarose beads functioned aseffectively as Ni-NTA Biacore sensor chips in protease selec-tion using monoclonal protein-phage. For this experiment,we used a 1:1 mixture of His6-FLEXI-phage and His6-WT-UBQ-phage, which were the products of two different phage-mids carrying different antibiotic resistance. The phagemixture was bound to the Ni-NTA agarose beads and treatedwith chymotrypsin. After various times, the beads were trea-ted with the serine-protease inhibitor PMSF, washed and theremaining phage were eluted with imidazole. The employ-ment of two different antibiotic resistances allowed the twopopulations of protein-phage to be gauged through this com-petitive proteolysis experiment as follows: phage rescued fromthe beads were split and used to infect E. coli, which werethen grown on separate agar plates containing the differentantibiotics. The number of colonies resulting on each plategave a measure of the relative fractions of His6-WT-UBQ-phage and His6-FLEXI-phage that remained intact andbound to the beads at each stage of the proteolysis experi-ment. This confirmed that UBQ-phage resisted proteolysisand that FELIX-phage was susceptible. It also demonstratedthat a phage population could be enriched for a stable pheno-type through protease selection.

We then turned to selection from the phage-displayedlibrary, His6-LIBRARY-UBQ-phage, using Ni-NTA agarosebeads as a support (32). Four rounds of selection were per-formed. Subsequent comparison of DNA sequences from thenaıve and selected libraries revealed a complete loss of the


parental AL7 mutant after selection, consistent with thisbeing a heavily destabilized mutant. In addition and intrigu-ingly, the selected library showed a clear preference for WTresidues in seven of the eight positions targeted in the initiallibrary. Several clones from the selected library were pickedand expressed. The resulting proteins were characterized byNMR and CD spectroscopy. All gave spectra consistent withfolding to native-like ubiquitin structures, and showed coop-erative thermal unfolding consistent with folding to uniquestructures. However, none of the selectants was more stablethan WT ubiquitin. This result is intriguing as it demon-strates that although protease selection can be used to rescuecompetently folded and stable mutants from a library ofhydrophobic core mutants, in the case of ubiquitin no super-stable variants emerge. Thus, whilst ubiquitin was a goodmodel for testing the concept of protease selection applied tocore-directed design (26,27,41) in some respects, in anotherit was a poor choice; in hindsight, we might not have expectedto stabilize ubiquitin. This is because the WT protein is ex-tremely stable and also highly conserved in nature. Further-more, to our knowledge, only one study has produced mutantsthat are more stable than the WT protein (41). This led us tosuggest that the WT ubiquitin sequences presents, or at leastis close to, the optimal solution for packing its hydrophobiccore (27).

Overall, using Ni-NTA agarose beads as an affinitymatrix to capture hexahistidine tagged protein-phage and inperforming protease selection is successful: well-folded ubi-quitin variants were recovered from a library of hydrophobiccore mutants and the destabilized parental clone—i.e., AL7

with its heavily compromised hydrophobic core—was lostthrough four rounds of protease selection. However, we didencounter problems with using Ni-NTA agarose beads.Namely, the whole process was labor-intensive and time con-suming. In addition, extensive washes were required afterbinding and protease challenge due to the well-documentedproblems associated with nonspecific binding of protein-phageto agarose and plastics. The former problem can be reducedto some extent by using the beads and other nickel-coated


surfaces to pull down intact phage after performing the selec-tion in solution. In addition, the necessity to determine phagetitres, and extensive plating out for each step to follow selec-tion was both laborious and time consuming. Hence, whenapplying our method, we recommend using SPR to examinemonoclonal protein-phage either side of protease selection,and for establishing the best buffer conditions for selection,followed by preparative-scale selections of phage-displayedlibraries using Ni-NTA agarose or other media.

V. STUDIES THAT BUILD ON THE ORIGINALMETHODS

Riechmann and Winter (42) describe an attempt to createnovel protein domains by complementing a fragment of aknown protein with randomly generated segments of polypep-tide. Specifically, random fragments of E. coli genomic DNAare generated by randomly primed PCR and fused to a geneencoding the N-terminal 36-residues of the cold shock proteinCspA, which correspond to the first three contiguous b-strands of the five-stranded b-barrel of CspA. The resultinglibrary is then subjected to phage display and proteolysis toselect stably folded chimeras. For this work, the groupswitches from the SIP-based method to using intact g3p andan N-terminal affinity tag as described by Finucane et al.(27,32)—except that the affinity tag is barnase—though theystill employ KM13 helper phage to rescue phage prior to pro-tease selection. Intriguingly, most of the selected chimerascontained genomic fragments in their orthodox readingframes. A number of the resulting proteins express and showcharacteristics of folded proteins: notably, weak but nonran-dom coil CD spectra; dispersion in 1D 1H NMR spectra; evi-dence for cooperative thermal unfolding; and stability withrespect to backbone-amide exchange. In a recent interestingdevelopment to the Riechmann and Winter experiment,Fischer et al. (43) combine two strands of an immunoglobulindomain with random fragments generated from humancDNA. In this case, one of the selected proteins, 2a6,


expresses well in E. coli; gives a CD indicative of mixeda- and b-structure; and forms a cooperatively folded, thermo-stable dimer. The selection of multimers using phage displayin this study and the foregoing work by Riechmann and Win-ter is interesting and is discussed to some extent in the morerecent paper.

Martin et al. (44) also use Proside to select for stabilizedvariants of a cold shock protein (CspB) with optimized sur-faces. They point out that 12 exposed surface residues differbetween a mesophilic Bs-CspB from Bacillus subtilis, and itsthermophilic counterpart Bc-CspB form Bacillus caldolyti-cus. On this basis, they randomize six of these sites to createa library of Bs-CspB variants. Using several rounds of pro-tease selection of varying stringency—namely, (1) in the pre-sence of 1.5M guanidine hydrochloride at 25�C and (2) at57.5�C and in a buffer of low ionic strength—mutants withhigher stabilities than the WT Bs-CspB are identified,including several that are significantly more stable thanthe thermophilic Bc-CspB. Interestingly, the selected resi-dues differ at all six positions from the thermophilic Bc-CspB, and at five of the six positions from the homologousTm-Csp from the hyperthermophilic Thermotoga maritima.The group has followed this work-up with site-directedmutagenesis and thermodynamic studies to unravel the ori-gins of the enhanced stabilities of the selected Bs-CspBmutants (45). Through these studies, Martin and Schmid(46) argue very convincingly that protein surfaces offer goodtargets for optimizing protein stability. Finally, the grouphas used Proside to engineer a thermostable g3p to aid theirselection studies.

In an interesting extension to his earlier studies withWinter, Kristensen (with colleagues) describes protease-basedselection on a library of barnase mutants (47). Library con-struction was guided by natural variations in the ribonu-clease observed by comparing the sequences of barnase andbinase. Bias towards specific amino-acid combinations wasobserved at four of the targeted sites, and at three of theseenrichments was with residues not present in either ribonu-clease. In addition, for some of the selectants improved


proteolytic stability and selection were not mirrored byimproved thermodynamic stability.

Bai and colleagues (35) have tested the approach of core-directed design further by attempting to convert a partiallyunfolded four-helix bundle, apocytochrome b562, into astably folded protein. Three residues from the anticipatedhydrophobic core were subjected to saturation mutagenesis.In addition, another core residue was changed to Trp toprovide a fluorescence probe, and a proximal residue wassubstituted with Arg to provide a specific, Arg-c protease site.Interestingly, after selection only one of the targeted sitesshowed any strong preference for hydrophobic residues.Nevertheless, the selected proteins were stable and amenableto high-resolution NMR and crystallographic studies, whichconfirmed four-helix-bundle structures. These results suggestthat core hydrophobic interactions alone may not dictate pro-tein stability and selection, although in this particular casethe location of the cutting site of the protease also appearsto influence selection.

Phage display and proteolysis have also been used toselect peptide sequences. For example, DeGrado and cowor-kers (34) combine peptide binding and protease treatment toselect new WW domain sequences from a phage-displayedlibrary. This work identified novel peptide-binding WWmotifssome of which showed cooperative unfolding, whilst others didnot. The approach has also been used in an attempt to selectprotease-resistant forms of a zinc-finger-based bba-structure(48). Although no monomeric and correctly folded selectantswere identified, somewhat curiously, peptides that formedamyloid-like assemblies were retrieved.

VI. SUMMARY

In conclusion, proteolysis can be used in phage display toselect proteins based solely on their ability to fold competentlyto stable structures; i.e., in the absence of any other more-traditional selectable function of the displayed protein suchas binding. Subtly different approaches for applying protease


selection in phage display have been presented independentlyby three groups (30–32). In addition, all of these groups and,importantly, others have followed up the early work withadventurous applications of the method in the areas of proteinengineering and design. For example, Finucane and Woolfson(27) and Bai and coworkers (35) have combined the methodwith core-directed design to engineer proteins from the insideout; Schmid and colleagues (44) have succeeded in engineer-ing proteins with improved stabilities by selection fromlibraries of surface mutants; Kristensen and coworkers (47)have retrieved variants of ribonucleases from a library basedon combining sequence features of barnase and binase; andWinter and colleagues (42,43) have created novel protein chi-meras by rescuing fragments of natural proteins with thoseencoded by randomly generated pieces of genomic DNA.Through these studies, the protease-selection method is nowfirmly established in the repertoire of phage-display experi-ments. It is hoped that it will find increased application inthe fields of engineering protein folding and stability and ofdesigning proteins de novo.

REFERENCES

1. Woolfson DN. Core-directed protein design. Curr Opin StructBiol 2001; 11:464–471.

2. Kuhlman B, Dantas G, Ireton GC, Varani G, Stoddard BL,Baker D. Design of a novel globular protein fold with atomic-level accuracy. Science 2003; 302:1364–1368.

3. Lazar GA, Handel TM. Hydrophobic core packing and proteindesign. Curr Opin Chem Biol 1998; 2:675–679.

4. Street AG, Mayo SL. Computational protein design. StructFold Des 1999; 7:R105–R109.

5. Saven JG. Combinatorial protein design. Curr Opin StructBiol 2002; 12:453–458.

6. Richards FM, Lim WA. An analysis of packing in the protein-folding problem. Quart Rev Biophys 1993; 26:423–498.


7. Betz SF, Raleigh DP, Degrado WF. De-novo protein design—from molten globules to native-like states. Curr Opin StructBiol 1993; 3:601–610.

8. Betz SF, Bryson JW, Degrado WF. Native-like and structu-rally characterized designed alpha-helical bundles. Curr OpinStruct Biol 1995; 5:457–463.

9. Hill RB, Raleigh DP, Lombardi A, Degrado NF. De novo designof helical bundles as models for understanding protein foldingand function. Acc Chem Res 2000; 33:745–754.

10. Lim WA, Sauer RT. Alternative packing arrangements in thehydrophobic core of lambda-repressor. Nature 1989; 339:31–36.

11. Lim WA, Sauer RT. The role of internal packing interactionsin determining the structure and stability of a protein. J MolBiol 1991; 219:359–376.

12. Lim WA, Farruggio DC, Sauer RT. Structural and energeticconsequences of disruptive mutations in a protein core.Biochemistry 1992; 31:4324–4333.

13. Axe DD, Foster NW, Fersht AR. Active barnase variants withcompletely random hydrophobic cores. Proc Natl Acad Sci USA1996; 93:5590–5594.

14. Uversky VN. Natively unfolded proteins: a point where biologywaits for physics. Prot Sci 2002; 11:739–756.

15. Kamtekar S, Schiffer JM, Xiong HY, Babik JM, Hecht MH.Protein design by binary patterning of polar and nonpolaramino-acids. Science 1993; 262:1680–1685.

16. Roy S, Helmer KJ, Hecht MH. Detecting native-like propertiesin combinatorial libraries of de novo proteins. Fold Des 1997;2:89–92.

17. Roy S, Ratnaswamy G, Boice JA, Fairman R, McLendon G,Hecht MH. A protein designed by binary patterning of polarand nonpolar amino acids displays native-like properties.J Amer Chem Soc 1997; 119:5302–5306.

18. Roy S, Hecht MH. Cooperative thermal denaturation ofproteins designed by binary patterning of polar and nonpolaramino acids. Biochemistry 2000; 39:4603–4607.


19. Wei YN, Liu T, Sazinsky SL, Moffet DA, Pelczer I, Hecht MH.Stably folded de novo proteins from a designed combinatoriallibrary. Prot Sci 2003; 12:92–102.

20. Wei YN, Kim S, Fela D, Baum J, Hecht MH. Solution structureof a de novo protein from a designed combinatorial library.Proc Natl Acad Sci USA 2003; 100:13270–13273.

21. Wei Y, Hecht MH. Enzyme-like proteins from an unselectedlibrary of designed amino acid sequences. Prot Eng Des Select2004; 17:67–75.

22. West MW, Wang WX, Patterson J, Mancias JD, Beasley JR,Hecht MH. De novo amyloid proteins from designed combina-torial libraries. Proc Natl Acad Sci USA 1999; 96:11211–11216.

23. Wang WX, Hecht MH. Rationally designed mutations convertde novo amyloid-like fibrils into monomeric beta-sheetproteins. Proc Natl Acad Sci USA 2002; 99:2760–2765.

24. Richardson JS, Richardson DC. Natural beta-sheet proteinsuse negative design to avoid edge- to-edge aggregation. ProcNatl Acad Sci USA 2002; 99:2754–2759.

25. Silverman JA, Balakrishnan R, Harbury PB. Reverse engi-neering the (beta=alpha) (8) barrel fold. Proc Natl Acad SciUSA 2002; 98:3092–3097.

26. Lazar GA, Desjarlais JR, Handel TM. De novo design of thehydrophobic core of ubiquitin. Prot Sci 1997; 6:1167–1178.

27. Finucane MD, Woolfson DN. Core-directed protein design. II.Rescue of a multiply mutated and destabilized variant ofubiquitin. Biochemistry 1999; 38:11613–11623.

28. Gu HD, Yi QA, Bray ST, Riddle DS, Shiau AK, Baker D. Aphage display system for studying the sequence determinantsof protein-folding. Prot Sci 1995; 4:1108–1117.

29. Chakravarty S, Mitra N, Queitsch I, Surolia A, Varadarajan R,Dubel S. Protein stabilization through phage display. FEBSLett 2000; 476:296–300.

30. Kristensen P,Winter G. Proteolytic selection for protein foldingusing filamentous bacteriophages. Fold Des 1998; 3:321–328.


31. Sieber V, Pluckthun A, Schmid FX. Selecting proteins withimproved stability by a phage-based method. Nat Biotechnol1998; 16:955–960.

32. Finucane MD, Tuna M, Lees JH, Woolfson DN. Core-directedprotein design. I. An experimental method for selecting stableproteins from combinatorial libraries. Biochemistry 1999; 38:11604–11612.

33. Matthews DJ, Wells JA. Substrate phage—selection of pro-tease substrates by monovalent phage display. Science 1993;260:1113–1117.

34. Dalby PA, Hoess RH, DeGrado WF. Evolution of binding affi-nity in a ww domain probed by phage display. Prot Sci 2000;9:2366–2376.

35. Chu R, Takei J, Knowlton JR, Andrykovitch M, Pei WH,Kajava AV, Steinbach PJ, Ji XH, Bai YW. Redesign of afour-helix bundle protein by phage display coupled withproteolysis and structural characterization by nmr and x-raycrystallography. J Mol Biol 2002; 323:253–262.

36. Fontana A, deLaureto PP, DeFilippis V, Scaramella E,Zambonin M. Probing the partly folded states of proteins bylimited proteolysis. Fold Des 1997; 2:R17–R26.

37. Clackson T, Winter G. Sticky feet-directed mutagenesis and itsapplication to swapping antibody domains. Nucleic Acids Res1989; 17:10163–10170.

38. Szabo A, Stolz L, Granzow R. Surface-plasmon resonance andits use in biomolecular interaction analysis (bia). Curr OpinStruct Biol 1995; 5:699–705.

39. Lasonder E, Schellekens GA, Welling GW. A fast and sensitivemethod for the evaluation of binding of phage clones selectedfrom a surface displayed library. Nucleic Acids Res 1994;22:545–546.

40. Rich RL, Myszka DG. Advances in surface plasmon resonancebiosensor analysis. Curr Opin Biotechnol 2000; 11:54–61.

41. Khorasanizadeh S, Peters ID, Roder H. Evidence for a three-state model of protein folding from kinetic analysis of ubiquitinvariants with altered core residues. Nat Struct Biol 1996; 3:193–205.


42. Riechmann L, Winter G. Novel folded protein domains gener-ated by combinatorial shuffling of polypeptide segments. ProcNatl Acad Sci USA 2000; 97:10068–10073.

43. Fischer N, Riechmann L, Winter G. A native-like artificial pro-tein from antisense DNA. Prot Eng Des Select 2004; 17:13–20.

44. Martin A, Sieber V, Schmid FX. In-vitro selection of highlystabilized protein variants with optimized surface. J MolBiol 2001; 309:717–726.

45. Martin A, Kather I, Schmid FX. Origins of the high stability ofan in vitro-selected cold-shock protein. J Mol Biol 2002; 318:1341–1349.

46. Martin A, Schmid FX. Evolutionary stabilization of the gene-3-protein of phage fd reveals the principles that govern thethermodynamic stability of two-domain proteins. J Mol Biol2003; 328:863–875.

47. Pedersen JS, Otzen DE, Kristensen P. Directed evolution ofbarnase stability using proteolytic selection. J Mol Biol 2002;323:115–123.

48. Koscielska-Kasprzak K, Otlewski J. Amyloid-forming peptidesselected proteolytically from phage display library. Prot Sci2003; 12:1675–1685.

49. Tuna M, Finucane MD, Vlachakis NGM, Woolfson DN. Pro-tease-based selection of stably folded proteins and proteindomains from phage display libraries. In: Clackson T, LowmanHB, eds. ‘Phage Display A Practical Approach’’. Oxford, 2004.


11

Identification of NaturalProtein–Protein Interactions

with cDNA Libraries

RETO CRAMERI, CLAUDIO RHYNER,and MICHAEL WEICHEL

Swiss Institute of Allergy and AsthmaResearch (SIAF), Davos, Switzerland

SABINE FLUCKIGER

BioVisioN Schweiz AG,Davos, Switzerland

ZOLTAN KONTHUR

Max Planck Institute of MolecularGenetics, Berlin, Germany

I. OVERVIEW

The genome projects provided us with a huge amount ofinformation at the DNA level and lead to the identification ofthousands of open reading frames. The demand for technolo-gies allowing the functional analysis of gene products is, there-fore, dramatically increased. Discovery and characterization of

415

interacting gene products, molecular recognition, and molecu-lar modeling became central to life sciences. Surface displaytechnology based on two pivotal concepts—physical linkagebetween genotype and phenotype and rescue of individualclones from large libraries by affinity selection—has the poten-tial to substantially contribute to functional genomics. Theexpansion of surface display technology in biosciences is facili-tated by the adaptability of the systems to high-throughputscreening formats for automated library handling. WhilerecombinantDNA techniques allow construction of highly com-plex molecular libraries, high-throughput screening allowsrapid exploration of molecular diversity using combinatorialmethods. These technologies are becoming increasingly impor-tant as molecular tools for the understanding of protein–protein interactions and for the generation of lead compounds,which, hopefully, will attract the business community to makeinvestments in this novel segment of biotechnology.

II. INTRODUCTION

All surface display technologies exploit the concept of linkingthe phenotype as a gene product displayed on a surface to itsgenetic information integrated into the host genome (1). Thisconcept is independent from the organism used and has beensuccessfully applied to construct large molecular libraries infilamentous phage (2–4), phagemids (5–7), lytic bacterio-phages (8,9), higher viruses (10,11) as well as prokaryotic(12–14), and eukaryotic (15,16) surface expression systems.When Smith (2) initially proposed the idea of phage displayin 1985, he suggested that selection of genes from cDNAlibraries could be one of the most significant applications ofthe technology. However, this potentially interesting area ofresearch has lagged behind, despite the impressive progressof phage display technology achieved during the last years.Among the over 2000 papers describing the use of phagedisplay available to date from the literature, only a few dealwith selection of cDNAs.

416 Crameri et al.

One of the reasons thereof is a direct consequence of thecapsid structure. Most phage or phagemid cloning vectors takeadvantage of the ability to assemble phage decorated withhybrid versions of the receptor protein pIII or the major coatprotein pVIII (17,18). This strategy has proven to be useful forthe N-terminal display of random peptide libraries (3,19–22),antibody fragments (23–26), and single proteins and proteindomains (27–30) which are directly fused to the coat proteins(31) or to truncated forms thereof in phagemid vectors (32).These approaches have been very successful because they allowdirect fusions of the gene products to be displayed to theN-terminus of the capsid proteins. However, the integrity ofthe C-terminus of pIII and pVIII is essential for efficient phageassembly and, therefore, the original vectors can only tolerateinsertion of foreignDNAat theN-terminus (5,9,33). This repre-sents the strongest limitation for the construction of cDNAdisplay libraries. The cDNA inserts encoding the C-terminusof proteins as obtained after poly(A) -priming and reverse tran-scription (34) always contains translation stop codons, whichprevent the synthesis of hybrid coat proteins (5,33,35). To over-come this limitation, several strategies, described in detailsbelow, have been devised. Fewer efforts have been invested intheuse of the remaining three capsidproteinspVI (36–38), pVII,and pIX (39), but their adequacy as vectors to display cDNAlibraries has not yet been tested extensively.

III. CLONING VECTORS

The basic idea of phage display technology consists in thesynthesis of a recombinant protein as fusion with a phage coatprotein, provided that the fusion does not interfere withphage infectivity or assembly. Historically, the first phagevectors allowing the fusion of polypeptides to pIII or pVIIIcontained the whole genome. According to the nomenclatureproposed by Smith (40), these phage vectors can be describedas type 3 and type 8 vectors (Fig. 1). The strongest limitationof these types of vectors is related to the short length of theinserts tolerated by the phage. The major coat protein pVIII

Protein–Protein Interactions with cDNA Libraries 417

can only tolerate very small inserts of about six to eight aminoacids between the N-terminal residues three and four (41,42).The pIII allows larger peptides and small proteins to be pre-sented as fusion between the export leader sequence anddomain D1 without dramatically affecting its function(19,43). Meanwhile, hybrid phage (type 33 and 88, Fig. 1)has been developed which contains both wild type and fusioncoat proteins integrated into the genome (44). The possibledrawback of these vectors consists in recombination eventsbetween the homologous wild type and fusion DNA regions,resulting in the loss of the information required for theproduction of fusion proteins (45).

Figure 1 Different mono- and multivalent M13-based phagesurface display systems. Vectors of the type 3, 8, 33 and 88 are mod-ified wild type phage. All other systems are phagemid vectors andrequire co-infection with wild type phage for assembly of infectivephagemid particles. See text for further explanations.

418 Crameri et al.

By combination of the best features of phage and plas-mids, new types of vectors termed phagemids were created(46,47). These vectors offer several advantages compared tofilamentous phage, such as easy preparation of high yieldsof dsDNA for cloning and sequencing, simple maintenanceas replicative plasmid form in bacteria, and adaptability torobot-assisted high-throughput screening technologies(48–50). Phagemids contain a bacterial and a phage originof replication, the phage packaging signal, antibiotic resis-tance genes for selection of transformants, and the geneencoding a coat protein used to generate fusions to be dis-played on phage surface. As a consequence thereof, they repli-cate in the host as plasmids and are able to be packaged in aphagemid particle, or recombinant phage, upon infection witha helper phage that provides the genes for the production ofthe structural, the packaging and assembly proteins neededfor phage morphogenesis. Sophisticated helper phage carriesmutations in the origin of replication or packaging sequences.Therefore, during replication, the phagemid genome is pack-aged more efficiently than the helper phage genome. Thebig advantage of phagemid over phage vectors consists inthe possibility of displaying not only small, but also largerpeptides (51), large molecules such as antibody fragments(52–54) and many other proteins including enzymes (55,56),enzyme inhibitors (57), and products of cDNA libraries(4,7,33,35,58–60). This becomes possible because the helperphage carries the full complement of capsid-encoding genes.As a competition during phagemid assembly, a mixture ofwild type and fusion coat protein can be incorporated intothe phage coat (vectors of type 3þ 3 and 8þ 8, Fig. 1). More-over, the number of fusion protein copies incorporated intothe recombinant phage particle (valency) can be influencedusing inducible promoters inserted in front of the truncatedcoat protein gene on the phagemid genome (61).

More recently, other phage coat proteins have beenexploited for the display of fusions including pVI (36) usedto display cDNA products as C-terminal fusions (type 6þ 6,Fig. 1). The coat proteins pVII and pIX have been used asfusion partners for the display of heavy and light chain


antibody fragments (39); however, these approaches have sofar been less commonly used. In addition to the mentioned‘‘standard’’ phage display vectors, other pIII-based systemsclassified as ‘‘phage two-hybrid systems’’ have been reported.Formally these vectors have been termed SAP for selectionand amplification of phage (62) and SIP for selectively infec-tive phage (63,64). In these systems, the fusion proteins areexpressed directly followed by the D2 and D3 domains of pIIIrendering all phage infection defective. Infectivity, e.g., theability of the phage to bind to the F-pilus and hence to infectEscherichia coli cells is restored by the selection target fusedto the pIII domain(s) D1 or D1 and D2. The SIP technologymay represent a powerful tool for rapid selection of protein–protein interactions (65) in spite of the few applicationsreported so far. Possible display strategies and vectors havebeen reviewed recently (66) (see also Chapter 2) and willnot be discussed here in further detail.

IV. DISPLAY OF cDNA LIBRARIES ONPHAGE SURFACE

Highly diverse display libraries have been constructed byfusing either genomic (67,68) or cDNA fragments (5,7)(35,36,58,60) to gene III or gene VIII of filamentous phage.In both cases, the display can be a challenge as the presenceof stop codons can hamper the generation of N-terminalfusions to the coat proteins, a direct consequence of the capsidstructure (33,48,50). Since the integrity of the C-terminus ofpIII and pVIII is considered essential for efficient phageassembly, insertions of foreign peptides can only be toleratedat the N-terminus. However, this problem has been alleviatedusing different strategies. Fusion of cDNAs to the C-terminusof the gene VI protein is compatible with phage propagationand packaging (36) as demonstrated in pilot experiments(37). The feasibility of this approach has been clearly demon-strated by the isolation of peroxisomal proteins from humancDNA libraries (38,69) and of a collagen-binding proteinfrom a Necator americanus cDNA library (70). Moreover,

420 Crameri et al.

Fuh and Sidhu (22) and Fuh et al. (71) have demonstratedthat, in contrast to the common belief, polypeptides fused tothe C-terminus of both the M13 pIII and pVIII coat proteinsare functionally displayed on the phage surface. The C-term-inal fusion approach, although not widely used until now,could be of considerable importance to phage display technol-ogy and would allow broad investigations of biological pro-blems, which are not suited for N-terminal display. Mainareas of interest in this field are the study of protein–proteininteractions requiring free C-termini and functional screen-ing of cDNA libraries.

More sophisticated approaches are based on the ability toseparate the gene III product of filamentous phage into itsfunctional domains: the N-terminal domains binding to theF pilus and mediating infection, and the C-terminal domainmorphologically involved in capping the trailing end of thefilament according to the vectorial polymerization model(72,73). Although cleavage of the gene III product into twoseparate functional entities is incompatible with phage propa-gation, infectivity can be restored by joining the segmentsthrough noncovalent protein–protein interactions (74,75).The so-called SIP technology (64,65) can be efficiently usedto screen cDNA libraries for selection of proteins that interactwith a target molecule as demonstrated in a few cases (76,77).

However, the most widely used systems for the construc-tion and screening of cDNA libraries displayed on phage sur-face involve an indirect fusion strategy where cDNA insertsfused to the 30 end of the Fos leucine zipper are coexpressedwith a truncated form of the gene III product decorated withthe Jun leucine zipper (5). The phagemid derived by modifica-tion of phagemid pComb 3 (78) and formally termed pJuFo(Fig. 2) has been widely used for the isolation of IgE-bindingmolecules from complex allergenic sources as reviewed else-where (7,48,79–81). Selective enrichment of IgE-bindingmole-cules from cDNA libraries constructed using mRNA fromAspergillus fumigatus (82), Malassezia furfur (83), peanut(84,85), Alternaria alternata (86), Cladosporium herbarum(87), Coprinus comatus (88), storage mites (89), and wheatgerm (90) yielded phage displaying hundreds of different


IgE-binding proteins. Interestingly, some of these structuresrepresent phylogenetically conserved proteins and share ahigh degree of sequence identity to their human counterparts.Human proteins, including manganese-dependent superoxidedismutase (91), acidic P2 ribosomal protein (92), and cyclopilin(93) could also be directly selected from a human lung cDNAlibrary displayed on the pJuFo surface using sera of patientssensitized to A. fumigatus as ligand (94,95), thus demonstrat-ing cross-reactivity to the environmental allergens (91,92).

Of course, filamentous phages and phagemids are not thevectors of choice for high-level expression of recombinantproteins. Therefore, all cDNAs isolated from phage surfacedisplay libraries need to be subcloned in high-level expressionvectors and transformed to a suitable host if relevant amountsof protein are required, for example, for clinical studies (96).In general, inserts subcloned from selected phagemids intohigh-level expression vectors are well expressed because, fordisplay on phage surface, the genetic information needs tobe transcribed and translated by E. coli.

Filamentous phage display systems, like any othercloning system, are not universal as they are subjected tobiological restrictions imposed by the host and by the codonusage of the cloned inserts. Possible serious biological

Figure 2 Genetic elements of the pJuFo phagemid and proposedpathway for the assembly of phage surface displayed cDNAlibraries. (Modified from Ref. 5.)

422 Crameri et al.

limitations derive from the characteristics of the phage lifecycle. Filamentous phage particles are released from the hostcell without breaking the integrity of the cell membrane. Theproteins, which assemble to form the capsid in the periplas-mic space, must therefore cross the lipid bilayer of the innermembrane. Therefore, any fusion peptide or protein with bio-chemical characteristics preventing transmembrane trans-port will not be integrated into the capsid. To be recognizedby the ligand used for selection, displayed proteins need toadapt a conformation able to interact with the ligand. Thechemical characteristics of the periplasmic environment,which affect the folding and stability of the recombinant pro-teins displayed, may influence the ability of hybrid coat pro-teins to interact with the ligand used for selection. SincecDNA libraries encode very diverse protein domains with dif-ferent biochemical properties, it is probable that a subset ofthese proteins or protein fragments will not be displayedand thus not be present in the surface display library. In addi-tion, cDNA libraries, like any other molecular library dis-played on phage surface, suffer from host-specific biologicallimitations related to restriction in codon usage, refoldingpathways, and potential toxicity of the expressed gene pro-ducts for the heterologous host.

However, phage surface display of cDNAs allows for thesurvey of very large libraries using the discriminative powerof affinity selection against homo- and heterogeneous ligands.Although the most successful applications of the pJuFo-basedcloning technology are related to IgE-binding molecules usingserum IgE from allergic patients as ligand, other successfulapplications have been reported. Examples are the mappingof protein–ligand interactions using whole genome phage dis-play libraries (67), construction of vectors for stable immobili-zation of multimeric recombinant proteins (97), and detailedanalysis of the C5a anaphylatoxin effector domain (98) fol-lowed by selection of a C5a receptor antagonist (99). Pereboevet al. (100) have used pJuFo to display adenovirus type 5 fiberknob as a tetrameric molecule able to bind to the coxsackie-virus-Ad receptor, demonstrating the versatile applicabilityof the cloning system. More recently, pJuFo has been used


to select autoantigens from human cDNA libraries derivedfrom patients suffering from vitiligo (101) and prostate cancer(102). In the first study, purified IgG from serum of vitiligopatients was used to screen a melanocyte cDNA-phage sur-face display library resulting in the discovery of the mela-nin-concentrating hormone receptor 1 (MCHR1) as a novelautoantigen related to this autoimmune disorder. Immuno-reactivity against the receptor was demonstrated in sera ofvitiligo patients using radiobinding assays. Among sera fromhealthy controls and from patients with other autoimmunediseases, no immunoreactivity to MCHR1 was found, indicat-ing a high disease specificity of autoantibodies raised againstthe receptor. In the second study, a cDNA library constructedfrom mRNA isolated from a lymph node metastasis of apatient suffering from hormone refractory prostate cancer(HRPC) was screened with purified autologous and hetero-logous IgG of patients suffering from prostate cancer. Sequen-cing of single clones after four rounds of biopanning yieldeddifferent cDNAs depending on the amount of IgG used forscreening, some of them corresponding to already known can-cer-associated antigens. These results, together with theisolation of colorectal-tumor-associated antigens from a pri-mary colorectal tumor cDNA library displayed on the surfaceas fusion to the gene VI protein (37), demonstrate the applic-ability of phage surface display for identification of cDNAexpression products in such diseases as cancer and autoim-mune disorders. The pJuFo vector was also used to clone pro-teins directly interacting with the cytoplasmic tail of themurine IgE-antigen receptor from a murine B cell cDNAlibrary displayed on phage surface (103). In contrast to theprevious examples, which used heterogeneous ligands, ahomogeneous synthetic 28 amino acid long peptide derivedfrom the cytoplasmic tail of IgE was used as selection bait.Among the inserts from 30 randomly chosen clones sequencedafter five rounds of biopanning, two carried cDNA fragmentscoding for the hematopoietic protein kinase 1 (HPK1). TheBIACORE measurements showed that HPK1 interacts invitro with the cytoplasmic tail of IgE as expected. The bindingof HPK1 to the cytoplasmic domain of IgE indicates the

424 Crameri et al.

existence of an isotype-specific signal transduction and mayrepresent a missing link to upstream regulatory elements ofHPK1 activation.

These examples clearly demonstrate that phage-displayedcDNA expression cloning can be a powerful tool for the isolationof unknown genes. The great advantage of cDNA surface dis-play compared to conventional lambda phage-based methodsis that in many cases the functional activity of a protein struc-ture can be used to select interaction partners together withthe genetic information required for their production. Thus,sequencing of the DNA of the integrated section of the phagegenome can readily elucidate the amino acid sequence of adisplayed gene product.

V. PROBLEMS ASSOCIATED WITH THEDISPLAY OF cDNA LIBRARIES ON PHAGESURFACE

A growing number of observations, published or not, indicatethat filamentous phage display technology is subjected toseveral limitations, some of these already discussed.Obviously, the quality of any cDNA library depends directlyon the quality of the cDNA ligated into the vector, which, inturn, is determined by the quality of the mRNA. Althoughthe methods for the isolation of mRNA available are quitereliable, oligo(dT)-priming might generate a high frequencyof truncated cDNAs through internal poly(A)-priming duringreverse transcription (104). Therefore, for genome wide geneidentification, reverse transcription should be done usinganchored oligo(dT) primers, which diminish the generationof truncated cDNAs caused by internal poly(A)-priming. Dur-ing the construction of a cDNA-phage surface display library,every step should be optimized to create the highest frequencyof potentially expressible full-length cDNA inserts. Transfor-mation with empty phagemids or phagemids containing shortinserts that have a growth advantage over large insert-containing phagemids may result in these undesirable clonesbecoming over-represented during library amplification.Avoiding overgrowth by defective clones is especially


important for large and highly heterogeneous cDNA expres-sion libraries.

Translational problems related to the codon usage mightbe alleviated by the use of hosts harboring genes encodinglimiting tRNA species like argU, ileY, and leuW to attain rea-sonable expression levels of proteins affected by rare codonusage (105). Like other prokaryotic-based expression systems,phage display may not be suitable for selection of proteinsthat require post-translational modifications (e.g. glycosyla-tion, phosporylation, etc.) or heterodimeric assembly for func-tional activity. Moreover, due to the absence of mammalianchaperonins, conformationally dependent structures maynot be efficiently expressed or refolded in these bacterialexpression systems and therefore not recognized by the ligandused for selection.

However, a successful screening does not only depend onbiological factors, but also on the biopanning strategy used(1,18). Selective enrichment of clones of interest becomesnecessary since phage surface display libraries contain largenumbers of cognate and uncognate phage. In a standardamplified library with a diversity of 108, each single clone ispresent in several thousand copies among a population of1012–1013 phage molecules. Therefore, the use of efficientselection and screening procedures is one of the key elementswhich determine the success of the combinatorial approach asdiscussed in details elsewhere (106). Phage display techni-ques require immobilization of the target protein to a solidsupport during biopanning. The immobilization process mustmaintain the target in a native or native-like conformation forphage selection (107). The commonly used method of proteinimmobilization through direct adsorption to plastic surfacesdenatures many proteins making them unsuitable targetsfor phage selection. Indirect immobilization of biotinylatedligands on streptavidin coated surfaces or chemical cross-linking to bifunctional resins (108) has been reported to bemore successful for the generation of native-like ligandsurfaces and should be considered whenever possible.

In spite of these limitations, phage display technologyhas significant advantages over other screening methods.

426 Crameri et al.

Compared to conventional bacterial or lambda phage-basedexpression systems, which are submitted to the same biologi-cal limitations, phage display technology enables rapid andselective enrichment of desired clones in small volumes usingminimal amounts of ligand molecules which can be, indeed, alimiting factor for selection.

VI. ADAPTABILITY OF PHAGE DISPLAYTO HIGH-THROUGHPUT SCREENINGTECHNOLOGY

The identification of the proteins produced in a given biologicalsystem started years ago with the discovery and improvementof recombinant DNA technologies that allowed controlledexpression of genes in many different hosts (109). However,cDNA-cloning technology including DNA sequencing can notbe directly used to study protein–protein interactions; thechallenge of functional genomics aimed to turn sequenceinformation into function. Estimates of the total number ofproteins resulting from transcription of the approximately35,000 human genes vary from 300,000 to millions (110), thusallowing for a much greater number of potential protein–protein interactions. Although many phenotypes can alreadybe pinpointed to their genetic origin through the sequence infor-mation of genome projects, many others remain un-known. Unfortunately, sequence information in itself is neithersufficient to provide significant knowledge of the underlyingmechanisms of life, nor of the biology of organisms. It rather pro-vides a sound basis and framework for further investigations.

The rapid identification of complex networks of interact-ing molecules in cells and tissues requires technologies thatprovide logistic and=or physical links between proteins andthe genes that encode them. The complex nature of molecularinteractions and the large numbers of diverse moleculesinvolved in biological processes require high-throughput tech-nology, allowing a sufficient degree of parallelization (50).New technologies for large scale analysis of genes andproteins have been devised such as differential display,RNA=DNA microarrays, and mass spectrometry (111) which,


however, suffer from the lack of a physical link betweensequence information and function. Phage display providesa physical link between genotype and phenotype (2,5,6) andallows the handling of large libraries based on the power ofaffinity selection (78). This physical link can easily becombined with a logistic protein–DNA link provided by robottechnology, enabling high-throughput picking and high-density arraying of single clones (50). These high-density arrayshave the advantage that each clone has a unique positiondefined by the coordinates on the microtiter plate and allowan unequivocal identification of each clone in later stages.

It has been shown that a human fetal brain cDNA expres-sion library can be screened in parallel for either DNA hybri-dization, protein expression, and for antibody screening in ahigh-density array format on filter membranes (112,113). Thisrobot-based high-throughput screening technology has beensuccessfully applied to phagemid libraries expressing complexallergen repertoires preselected with serum IgE of allergicindividuals (48,79,81,95). The potential of the combination ofcDNA-phage surface display with selection for specific inter-action by functional screening and robotic technology isillustrated by the isolation of more sequences potentiallyencoding IgE-binding proteins than postulated from Westernblot analysis using extracts derived from raw material ofcomplex allergenic sources (114). Moreover, robot-based high-throughput screening technology has been applied to recombi-nant antibody arrays displayed on phage surface to detectantibody–antigen interactions (49). Therefore, high-through-put screening technology applied to complex surface displayedcDNA libraries will play an important role in the postgeno-mic era by identifying potential ligands against large numbersof diverse molecules expressed by cell cultures, tissues, ororganisms.

VII. CONCLUSIONS

The major challenge in the postgenomic era is to turnsequence information into function. The keymolecular players

428 Crameri et al.

in cells and tissues, which are instrumental for the functioningof an organism, are the proteins. Built up from 20 differentamino acids encoded by the DNA of a limited number of genes,proteins are produced through complex translational andpost-translational pathways generating a great deal of func-tional structures. This diversity enables complex networksof molecular interactions governing the functioning of anorganism. The characterization of large numbers of genes,their expression patterns, and protein interactions demandsthe use of high-throughput technologies able to link infor-mation, deposited in the genome, to function exerted by theproteins themselves.

Phage surface display of cDNAs as a biological approachlinking genotype and phenotype, although subjected to intrin-sic biological limitations, can be used for the efficient identifi-cation of gene products based on protein–protein interactions.Thus, the technology has the potential for substantially con-tributing to rapid developments in functional genomics. Thebasic knowledge accumulated from successful and unsuccess-ful applications of cDNA-phage surface display will improveour understanding of the biological limitations of the systemscurrently used, and thus will help further improve thetechnology.

ACKNOWLEDGMENTS

We are grateful to Prof. K. Blaser and Prof. H. Lehrach forcontinuous support and encouragement. This work wassupported by the Swiss National Science Foundation grant31.63382.00.

REFERENCES

1. Borrebaeck CA. Tapping the potential of molecular librariesin functional genomics. Immunol Today 1998; 19:524–527.

2. Smith GP. Filamentous fusion phage: novel expressionvectors that display cloned antigens on the virion surface.Science 1985; 228:1315–1318.




5. Crameri R, Suter, M. Display of biologically active proteins onthe surface of filamentous phages: a cDNA cloning system forselection of functional gene products linked to the geneticinformation responsible for their production. Gene 1993;137:69–75.

6. Barbas CF III, Kang AS, Lerner RA, Benkovic SJ. Assemblyof combinatorial antibody libraries on phage surfaces: thegene III site. Proc Natl Acad Sci USA 1991; 88:7978–7982.

7. Rhyner C, Kodzius R, Crameri R. Direct selection of cDNAsfrom filamentous phage surface display libraries. Potentialand limitation. Curr Pharm Biotechnol 2002; 3:13–21.

8. Mikawa YG, Maruyama IN, Brenner S. Surface display ofproteins on bacteriophage lamda heads. J Mol Biol 1996;262:21–30.

9. Castagnoli L, Zucconi A, Quondam M, Rossi M, Vaccaro P,Panni S, Paoluzi S, Santonico E, Denti L, Cesareni G. Alter-native bacteriophage display systems. Combin Chem HighThroughput Screen 2001; 4:121–133.

10. Grabherr R, Ernst W. The baculovirus expression system as atool for generating diversity by viral surface display. CombinChem High Throughput Screen 2001; 4:185–192.

11. Grabherr R, Ernst W, Oker-Blom C, Jones I. Developments inthe use of baculoviruses for the surface display of complexeukaryotic proteins. Trends Biotechnol 2001; 19:231–236.

12. Hansson M, Samuelson P, Gunneriusson E, Stahl S. Surfacedisplay on Gram positive bacteria. Combin Chem HighThroughput Screen 2001; 4:171–184.

13. Stahl S, Uhlen M. Bacterial surface display: trends andprogress. Trends Biotechnol 1997; 15:185–192.

14. Georgiou G, Stathopoulos C, Daugherty PS, Nayak AR,Iverson BL, Curtiss R III. Display of heterologous proteinson the surface of microorganisms. From the screening of

430 Crameri et al.

combinatorial libraries to live recombinant vaccines. NatBiotechnol 1997; 15:29–34.

15. Boder ET, Wittrup KD. Yeast surface display for directed evo-lution of protein expression, affinity, and stability. MethodsEnzymol 2000; 328:430–444.

16. Yeung YA, Wittrup KD. Quantitative screening of yeastsurface-displayed polypeptide libraries by magnetic beadcapture. Biotechnol Prog 2002; 18:212–220.

17. Mead DA, Kemper B. Chimeric single-stranded DNA phage-plasmid cloning vectors. Biotechnology (NY) 1988; 10:85–102.

18. Kay BK, Winter J, McCafferty J. Phage Display of Peptidesand Proteins. A Laboratory Manual. San Diego, CA:Academic Press Inc, 1996.

19. Devlin JJ, Panganiban LC, Devlin PE. Random peptidelibraries: a source of specific protein-binding molecules.Science 1990; 249:404–406.

20. Barrett RW, Cwirla SE, Ackerman MS, Olson AM, Peters EA,Dower WJ. Selective enrichment and characterization of highaffinity ligands from collections of random peptides onfilamentous phage. Anal Biochem 1992; 204:357–364.

21. Cabilly S. The basic structure of filamentous phage and itsuse in the display of combinatorial peptide libraries. MolBiotechnol 1999; 12:143–148.

22. Fuh G, Sidhu SS. Efficient phage display of polypeptidesfused to the carboxy-terminus of the M13 gene-3 minor coatprotein. FEBS Lett 2000; 480:231–234.

23. BurtonDR,BarbasCFIII,PerssonMA,KoenigS,ChanockRM,Lerner RA. A large array of human monoclonal antibodies totype 1 human immunodeficiency virus from combinatoriallibraries of asymptomatic seropositive individuals. Proc NatlAcad Sci USA 1991; 88:10134–10137.

24. Fisch I, Kontermann RE, Finnern R, Hartley O,Soler-Gonzalez AS, Griffiths AD, Winter G. A strategy of exonshuffling for making large peptide repertoires displayed onfilamentous bacteriophage. Proc Natl Acad Sci USA 1996; 93:7761–7766.


25. Sblattero D, Bradbury A. Exploiting recombination insingle bacteria to make large phage antibody libraries. NatBiotechnol 2000; 18:75–80.

26. Liu B, Huang L, Sihlbom C, Burlingame A, Marks JD.Towards proteome-wide production of monoclonal antibodyby phage display. J Mol Biol 2002; 315:1063–1073.

27. Zucconi A, Panni S, Paoluzi S, Castagnoli L, Dente L,Cesareni G. Domain repertoires as a tool to derivate proteinrecognition rules. FEBS Lett 2000; 480:49–54.

28. Schiffer C, Ultsch M, Aalsh S, Somers W, de Vos AM,Kossiakoff A. Structure of a phage display-derived variantof human growth hormone complexed to two copies of extra-cellular domain of its receptor: evidence for strong structuralcoupling between receptor-binding sites. J Mol Biol 2002;316:277–289.

29. Fazi B, Cope MJ, Douangamath A, Ferracuti S, Schiriwitz K,Zucconi A, Drubin DG, Wilmanns M, Cesareni G, CastagnoliL. Unusual binding properties of the SH3 domain of the yeastactin-binding protein Abp1: structural and functional analy-sis. J Mol Biol 2002; 277:5290–5298.

30. Reichmann L, Winter G. Novel folded protein domains gener-ated by combinatorial shuffling of polypeptide segments. ProcNatl Acad Sci USA 2000; 97:10068–10073.

31. Smith GP. Filamentous phages as cloning vectors. Biotech-nology (NY) 1988; 10:61–83.

32. Sidhu SS. Engineering M13 for phage display. Biomol Eng2001; 18:57–63.

33. Crameri R, Hemmann S, Blaser K. pJuFo: a phagemid fordisplay of cDNA libraries on phage surface suitable forselective isolation of clones expressing allergens. Adv MedExp Biol 1996; 409:103–110.

34. Nam DK, Lee S, Zhou G, Cao X, Wang C, Clark T, Chen J,Rowley JD, Wang SM. Oligo(dT) priming generates a highfrequency of truncated cDNAs through internal poly(A) prim-ing during reverse transcription. Proc Natl Acad Sci USA2002; 99:6152–6156.

432 Crameri et al.

35. Crameri R, Jaussi R, Menz G, Blaser K. Display of expressionproducts of cDNA libraries on phage surfaces. A versatilescreening system for selective isolation of genes by specificgene-product=ligand interaction. Eur J Biochem 1994;226:53–58.

36. Jespers LS, Messens JH, De Keyser A, Eeckhout D, Van denBrande Gansemans YG, Lauwereys MJ, Vlasuk GP,Stanssens PE. Surface expression and ligand-based selectionof cDNAs fused to filamentous phage gene VI. Biotechnology(NY) 1995; 13:378–382.

37. Hufton SE, Moerkerk PT, Meulemans EV, de Bruine A,Arends JW, Hoogenboom HR. Phage display of cDNA reper-toires: the pVI display system and its applications for theselection of immunogenic ligands. Immunol Methods 1999;231:39–51.

38. Amery L, Mannaerts GP, Subramani S, Van Veldhoven PP,Fransen M. Identification of a novel human peroxisomal2,4-dienoyl-CoA reductase related protein using the M13phage protein VI phage display technology. Combin ChemHigh Throughput Screen 2001; 4:545–552.

39. Gao C, Mao S, Lo CH, Wirsching P, Lerner RA, Janda KD.Making artificial antibodies. A format for phage display ofcombinatorial heterodimeric arrays. Proc Natl Acad SciUSA 1999; 96:6025–6030.

40. Smith GP, Scott JK. Libraries of peptides and proteinsdisplayed on filamentous phage. Meth Enzymol 1993; 217:228–257.

41. Greenwood J, Willis AE, Perham RN. Multiple display offoreign peptides on a filamentous bacteriophage. Peptidesfrom Plasmodium falciparum circumsporozoite protein asantigens. J Mol Biol 1991; 220:821–827.

42. Petrenko VA, Smith GP, Gong X, Quinn T. A library of organiclandscapes on filamentous phage. Protein Eng 1996; 9:797–801.



44. Haaparanta T, Huse WD. A combinatorial method for con-structing libraries of long peptides displayed by filamentousphage. Mol Divers 1995; 1:39–52.

45. Bonnycastle LL, Mehroke JS, Rashed M, Gong X, Scott JK.Probing the basis of antibody reactivity with a panel ofconstrained peptide libraries displayed by filamentous phage.J Mol Biol 1996; 258:747–762.

46. Mead DA, Kemper B. Chimeric single-stranded DNA phage-plasmid cloning vectors. Biotechnology 1988; 10:85–102.

47. Larocca D, Burg MA, Jensen-Pegakes K, Ravey EP, GonzalesAM, Baird A. Evolving phage vectors for cell targeted genedelivery. Curr Pharm Biotechnol 2002; 3:45–57.

48. Crameri R, Kodzius R. The powerful combination of phagesurface display of cDNA libraries and high throughputscreening. Combin Chem High Throughput Screen 2001;4:145–155.

49. de Wildt RTM, Mundy CR, Gorick BD, Tomlinson IM.Antibody arrays for high-throughput screening of antibody–antigen interactions. Nat Biotechnol 2000; 18:989–994.

50. Walter G, Konthur Z, Lehrach H. High-throughput screeningof surface displayed gene products. Combin Chem HighThroughput Screen 2001; 4:193–205.

51. Wrighton NC, Farrell FX, Chang R, Kashyap AK, BarboneFP, Mucahy LS, Johnson DL, Barrett RW, Jolliffe LK, DowerWJ. Small peptides as potent mimetics of the proteinhormone erythropoietin. Science 1996; 273:458–464.

52. Rader C, Barbas CF III. Phage display of combinatorial anti-body libraries. Curr Opin Biotechnol 1997; 8:503–508.

53. Winter G. Making antibody and peptide ligands by repertoireselection technologies. J Mol Recognit 1998; 11:126–127.

54. Sblattero D, Lou J, Marzari R, Bradbury A. In vivo recombi-nation as a tool to generate molecular diversity in phage anti-body libraries. J Biotechnol 2001; 74:303–315.

55. Soumillion P, Jespers L, Bouchet M, Marchand-Brynaert J,Sartiaux P, Fastrez J. Phage display of enzymes and in vitro

434 Crameri et al.

selection for catalytic activity. Appl Biochem Biotechnol 1994;47:175–189.

56. Forrer P, Jung S, Pluckthun A. Beyond binding: using phagedisplay to select for structure, folding and enzymatic activityin proteins. Curr Opin Struct Biol 1999; 9:514–520.

57. Rottgen P, Collins J. A human pancreatic secretory trypsininhibitor presenting a hypervariable highly constrained epi-tope via monovalent phagemid display. Gene 1995; 164:243–250.

58. Dunn IS. Phage display of proteins. Curr Opin Biotechnol1996; 7:547–553.

59. Kemp EH, Waterman EA, Hawes BE, O’Neill K,Gottumukkala RV, Gawkrodger DJ, Weetman AP, WatsonPF. The melanin-concentrating hormone receptor 1, a noveltarget of autoantibody responses in vitiligo. J Clin Invest2002; 109:923–930.

60. Crameri R, Achatz G,Weichel M, Rhyner C. Direct selection ofcDNAs by phage display.MethodsMol Biol 2002; 184:461–469.

61. Huang W, McKevitt M, Palzkill T. Use of the arabinosep(bad) promoter for tightly regulated display of proteins onbacteriophage. Gene 2000; 251:187–197.

62. Duenas M, Borrebaeck CA. Clonal selection and amplificationof phage displayed antibodies by linking antigen recognitionand phage replication. Biotechnology (NY) 1994; 12:999–1002.

63. Spada S, Krebber C, Pluckthun A. Selectively infective phage(SIP). Biol Chem 1997; 378:445–456.

64. Arndt KM, Jung S, Krebber C, Pluckthun A. Selectively infec-tive phage technology. Meth Enzymol 2000; 328:364–388.

65. Jung S, Arndt KM, Muller KM, Pluckthun A. Selectivelyinfective phage (SIP) technology: scope and limitations.J Immunol Meth 1999; 231:93–104.

66. Irving MB, Pan O, Scott JK. Random-peptide libraries andantigen-fragment libraries for epitope mapping and thedevelopment of vaccines and diagnostics. Curr Opin ChemBiol 2001; 5:314–324.


67. Palzkill T, Huang W, Weinstock GM. Mapping protein–ligandinteractions using whole genome phage display libraries.Gene 1998; 222:79–83.

68. Jacobsson K, Frykberg L. Shotgun phage display cloning.Combin Chem High Throughput Screen 2001; 4:138–143.

69. Fransen M, Van Veldhoven PP, Bubramani S. Identificationof peroxisomal proteins by using M13 phage protein VIdisplay: molecular evidence that mammalian peroxisomescontain a 2,4-dienoyl-CoA reductase. Biochem J 1999; 340:561–568.

70. Viaene A, Carb A, Meiring M, Pritchard D, Deckmyn H.Identification of a collagen-binding protein from Necatoramericanus by using a cDNA-expression phage displaylibrary. J Parasitol 2001; 87:619–625.

71. Fuh G, Pisabarro MT, Li Y, Quan C, Lasky LA, Sidhu SS.Analysis of PDZ domain–ligand interactions using carboxy-terminal phage display. J Biol Chem 2000; 275:21486–21491.

72. Chang CN, Model P, Blobel G. Membrane biogenesis: cotran-slational integration of the bacteriophage f1 coat protein intoan Escherichia coli membrane fraction. Proc Natl Acad SciUSA 1979; 76:1251–1255.

73. Stengele I, Bross P, Graces S, Giray I, Rasched I. Dissectionof functional domains in phage fd adsorption protein. Discri-mination between attachment and penetration sites. J MolBiol 1990; 5:143–149.

74. Gramatikoff K, Georgiev O, Schaffner W. Direct interactionrescue, a novel filamentous phage technique to studyprotein–protein interactions. Nucleic Acid Res 1994; 22:5761–5762.

75. Krebber C, Spada S, Desplancq D, Krebber A, Ge L,Pluckthun A. Selectively-infective phage (SIP): a mechanisticdissection of a novel in vivo selection for protein–ligand inter-actions. J Mol Biol 1997; 268:607–618.

76. Gramatikoff K, Schaffner W, Georgiev O. The leucine zipperof c-Jun binds to ribosomal protein L18a: a role in Jun proteinregulation? Biol Chem Hoppe Seyler 1995; 376:321–325.

436 Crameri et al.

77. Hottiger M, Gramatikoff K, Georgiev O, Chaponnier C,Schaffner W, Hubscher U. The large subunit of HIV-1 reversetranscriptase interacts with beta-actin. Nucleic Acids Res1995; 23:736–741.

78. Barbas CF III, Lerner RA. Combinatorial immunoglobulinlibraries on the surface of phage (Phabs): rapid selection ofantigen-specific Fabs. Meth Compan Meth Enzymol 1991;2:119–124.

79. Crameri R, Walter G. Selective enrichment and high-throughput screening of phage surface-displayed cDNAlibraries from complex allergenic systems. Combin ChemHigh Throughput Screen 1999; 2:63–72.

80. Appenzeller U, Blaser K, Crameri R. Phage display as a toolfor rapid cloning of allergenic proteins. Arch Immunol TherExper 2001; 49:19–25.

81. Crameri R. High throughput screening: a rapid way to recom-binant allergens. Allergy 2001; 54(suppl 67):30–34.

82. Crameri R. Molecular cloning of Aspergillus fumigatus aller-gens and their role in allergic bronchopulmonary aspergillo-sis. Chem Immunol 2002; 71:73–93.

83. Lindborg M, Magnusson CGM, Zagari A, Schmidt M,Scheyinus A, Crameri R, Withley P. Selective cloning of aller-gens from the skin colonizing yeast Malassezia furfur byphage surface display technology. J Invest Dermatol 1999;113:156–161.

84. Kleber-Janke T, Crameri R, Appenzeller U, Schlaak M,Becker WM. Selective cloning of peanut allergens, includingprofilin and 2S albumin, by phage display technology. IntArch Allergy Immunol 1999; 119:265–274.

85. Kleber-Janke T, Crameri R, Scheurer S, Vieths S, BeckerWM. Patient-tailored cloning of allergens by phage display:peanut (Arachis hypogaea) profilin, a food allergen derivedfrom a rare mRNA. J Chromatogr B Biomed Sci Appl 2001;756:295–305.

86. Weichel M, Schmid-Grendelmeier P, Fluckiger S,Breitenbach M, Blaser K, Crameri R. Nuclear transport fac-


tor 2 represents a novel cross-reactive fungal allergen.Allergy 2003; 58:198–206.

87. Weichel M, Schmid-Grendelmeier P, Rhyner C, Achatz G,Blaser K, Crameri R. IgE-binding and skin test reactivityto hydrophobin HCh-1 from Cladosporium herbarum, thefirst allergenic cell wall component of fungi. Clin Exp Allergy2003; 33: 72–77.

88. Brander KA, Borbley P, Crameri R, Pichler WJ, Helbling A.IgE-binding proliferative responses and skin test reactivityto Cop c 1, the first recombinant allergen for the basidiomy-cete Coprinus comatus. J Allergy Clin Immunol 1999; 104:630–636.

89. Eriksson TLJ, Rasool O, Huecas S, Whitley P, Crameri R,Appenzeller U, Gafvelin G, van Hage-Hamsten M. Cloningof three new allergens from the dust mite Lepidoglyphusdestructor using phage surface display technology. EurJ Biochem 2001; 266:287–294.

90. Rozynek P, Sander I, Appenzeller U, Crameri R, Baur X,Clarke T, Brunig B, Raulf-HeimsothM.BPIS—an IgE-bindingwheat protein. Allergy 2002; 57:463.

91. Crameri R, Faith A, Hemmann S, Jaussi R, Ismail C, Menz G,Blaser K. Humoral and cell-mediated autoimmunity in allergyto Aspergillus fumigatus. J Exp Med 1996; 184:265–270.

92. Mayer C, Appenzeller U, Seelbach H, Achatz G, Oberkofler H,Breitenbach M, Blaser K, Crameri R. Humoral and cell-mediated autoimmune reactions to human ribosomal P2

protein in individuals sensitized to Aspergillus fumigatus P2

protein. J Exp Med 1999; 189:1507–1512.

93. Fluckiger S, Fijten H, Whitley P, Blaser K, Crameri R.Cyclophilins, a new family of cross-reactive allergens. Eur JImmunol 2002; 32:10–17.

94. Appenzeller U, Mayer C, Menz G, Blaser K, Crameri R.IgE-mediated reactions to autoantigens in allergic diseases.Int Ach Allergy Immunol 1999; 118:193–196.

95. Crameri R, Kodzius R, Konthur Z, Lehrach H, Blaser K,Walter G. Tapping allergen repertoires by advanced cloningtechnologies. Int Arch Allergy Immunol 2001; 124:43–47.

438 Crameri et al.

96. Schmid-Grendelmeier P, Crameri R. Recombinant allergensfor skin testing. Int Arch Allergy Immunol 2001; 125:96–111.

97. Grob P, Baumann S, Ackermann M, Suter M. A system forstable indirect immobilization of multimeric recombinantproteins. Immunotechnology 1988; 4:155–165.

98. Hennecke M, Kola A, Baensch M, Wrede A, Klos A, BautschW, Kohl J. A selection system to study C5a–C5a-receptorinteractions: phage display of a novel C5a anaphylatoxin,Fos-C5aAla27. Gene 1997; 184:263–272.

99. Heller T, Hennecke M, Baumann U, Gessner JE, zuVilsendorf AM, Baensch M, Boulay F, Kola A, Klos A,Bautsch W, Kohl J. Selection of a c5a antagonist from phagelibraries attenuating the inflammatory response in immunecomplex disease and ischemia=reperfusion injury. J Immunol1999; 163:985–994.

100. Pereboev A, Pereboeva L, Curiel DT. Phage display of adeno-virus type 5 fiber knob as a tool for specific ligand selectionand validation. J Virol 2001; 75:7107–7113.

101. Kemp EH, Waterman EA, Hawes BE, O’Neill K,Gottumukkala RV, Gawkrodger DJ, Weetman AP, WatsonPF. The melanin-concentrating hormone receptor 1, a noveltarget of autoantibody responses in vitiligo. J Clin Invest2002; 109:993–998.

102. Fossa A, Alsoe L, Crameri R, Funderud S, Gaudernack G,Smeland EB. Serological cloning of cancer=testis antigensexpressed in prostate cancer using cDNA phage surfacedisplay. Cancer Immunol Immunother 2004; 53:431–438.

103. Geisberger R, Prlic M, Achatz-Straussberger G, OberndorferI, Luger E, Lamers M, Crameri R, Appenzeller U, WienandsJ, Breitenbach M, Ferreira F, Achatz G. Phage display basedcloning of proteins interacting with the cytoplasmic tail ofmembrane immunoglobulins. Dev Immunol 2002; 9:127–134.

104. Nam DK, Lee S, Zhou G, Cao X, Wang C, Clark T, Chen J,Rowley JD, Wang SM. Oligo(dT) primer generates a high fre-quency of truncated cDNAs through internal poly(A) primingduring reverse transcription. Proc Natl Acad Sci USA 2002;99:6152–6156.


105. Kleber-Janke T, Becker WM. Use of modified BL21(DE3)Escherichia coli cells for high-level expression of recombinantpeanut allergens affected by poor codon usage. Protein ExprPurif 2000; 19:419–424.

106. Levitan B. Stochastic modelling and optimization of phagedisplay. J Mol Biol 1998; 277:895–916.

107. Suter M, Foti M, Ackermann M, Crameri R. In: Kay BK,Winter J, McCafferty J, eds. Phage Display of Peptides andProteins. A Laboratory Manual. San Diego: Academic PressInc, 1996:195–214.

108. Chernukhin IV, Klenova EM. A method of immobilization onthe solid support of complex and simple enzymes retainingtheir activity. Anal Biochem 2000; 280:178–181.

109. Baneyx F. Recombinant protein expression in Escherichiacoli. Curr Opin Biotechnol 1999; 10:411–421.

110. Li M. Application of display technology in protein analysis.Nat Biotechnol 2000; 18:1251–1256.

111. MacCoss MJ, Yates JR III. Proteomics: analytical tools andtechniques. Curr Opin Clin NutrMetab Care 2001; 4:369–375.

112. Bussow K, Nordhoff E, Lubbert C, Lehrach H, Walter G. Ahuman cDNA library for high-throughput protein expressionscreening. Genomics 2000; 65:1–8.

113. Walter G, Bussow K, Lueking A, Glokler J. High-throughputprotein arrays: prospects for molecular diagnostics. TrendsMol Med 2002; 8:250–253.

114. Kodzius R, Rhyner C, Konthur Z, Buczek D, Lehrach H,Walter G, Crameri R. Rapid identification of allergen-encod-ing cDNA clones by phage display and high-density arrays.Comb Chem High Throughput Screen 2003; 6:147–153.

440 Crameri et al.

12

Mapping Protein FunctionalEpitopes

SARA K. AVRANTINIS and GREGORY A. WEISS

Department of Chemistry, University ofCalifornia, Irvine, California, U.S.A.

I. INTRODUCTION

Essentially all events in biology require one molecule bindingto another. With molecular recognition at the heart of biology,a number of phage-display-based techniques have been devel-oped to dissect the details governing binding events. Thischapter reviews phage-display applications that provide scal-pels and microscopes to explore molecular recognition. Thesetechniques capitalize upon the advantages of the phage-display format—rapid production of protein variants, simpleprotein purification, and the potential for powerful selections.

Modifying the primary structure of a protein can explorethe intricate relationship between the primary sequence of

441

amino acids and the properties of the resultant protein,including shape, stability, and activity. Mutagenesis of speci-fic protein residues has proven invaluable in probing thecontributions of individual amino acid side chains to suchproperties. For example, the seminal chemical syntheses ofproteins with modified side chains by Hodges and Merrifield(1) provided a detailed understanding of the catalytic mecha-nism of RNase A. In another now classic experiment, Fershtet al. (2) elegantly demonstrated the power of oligonucleo-tide-directed mutagenesis to measure hydrogen bondstrengths within proteins. These experiments set the stagefor the emergence of high throughput techniques for proteinmodification and dissection using phage-displayed proteins.

The goal of the experiments described here is to identifythe side chains and functional groups responsible for a parti-cular protein function, often binding to a specific target.Quantitation of the energetic contributions made by specificside chains and functional groups is a welcome bonus madepossible by some phage-display techniques. For the examina-tion of receptor–ligand interactions, protein libraries can begenerated by either random (3–5) or site-specific mutations.Several options are available to install mutations site specifi-cally. Saturation mutagenesis at specific sites replaces a par-ticular side chain with all 20 naturally occurring amino acids(6,7). Substitution with a few carefully chosen amino acids, onthe other hand, can provide a thorough portrait of the contri-butions made by an individual side chain and potentiallyenable quantitative analysis of the data. In addition, limitedmutagenesis can allow for coverage of more protein residuesin a single library.

Noncovalent binding interactions between receptorsand ligands are mediated by directly contacting residues,which form a structural epitope. Biophysical techniques, suchas x-ray diffraction and multidimensional NMR, can revealsuch binding contacts. The structure and structural epitopealone, however, do not tell the whole story of how a proteinworks. Within the contact area, a subset of residues maycontribute the majority of binding energy or protein function-ality. Such residues can form key hydrogen bonds, salt

442 Avrantinis and Weiss

bridges, dipole–dipole, and hydrophobic interactions withinthe binding interface. Residues with energetically favorablecontacts compose the functional epitope of a binding protein(8). A tightly clustered functional epitope resembling a cross-section of protein structure (i.e., a hydrophobic center sur-rounded by hydrophilic groups) has been termed a ‘‘hot spot’’of binding energy (9–11). Identification of functional epitopesrequires quantification of individual side chain contributionsto protein function, a task eminently suited for phage-displaytechniques.

The functional epitope ultimately reveals how a proteinworks. However, understanding protein function also requiresidentification of residues that position the functional epitopeside chains (‘‘second sphere residues’’) (12) and even thirdsphere residues, which may influence protein activity. In addi-tion, thorough understanding of protein structure anddynamics includes identification of residues critical to proteinfolding and stability. Phage-display techniques can identifythese important residues.

This chapter surveys experiments demonstrating thescope of functional epitope mapping by phage-display techni-ques. The examples cited are not meant to be comprehensiveand the authors apologize in advance for not having the spaceto include all possible examples of functional epitope mappingby phage display. In general, this chapter focuses upon tech-niques with combinatorial, yet systematic, variation of speci-fic side chains. To introduce the topic, a few examples of singlepoint mutagenesis, both with and without phage display, willbe described.

II. SINGLE POINT ALANINE MUTAGENESIS

Alanine substitutions truncate amino acid side chains at theb-carbon. Thus, each alanine mutation explores the functionalimportance of side chain atoms past the b-carbon. Alanine istypically chosen because the substitution nullifies the aminoacid side chain without introducing additional conformationalflexibility into the protein backbone that might result from aglycine mutation. Alanine scanning mutagenesis, systematic

Mapping Protein Functional Epitopes 443

replacement of individual wild-type residues with alanine, is aparticularly useful technique for the identification of func-tional epitopes and consequently for uncovering biologicalinsight. For example, alanine scanning revealed that humangrowth hormone (hGH) binds to hGH-binding protein (hGHbp)with a remarkably compact patch of 8 out of 31 resides buriedat the interface (8). The hGHbp hotspot, also determined byalanine scanning, consists of a patch of residues with comple-mentary size and functional groups (9). In another example,researchers examined the first committed step in HIV infec-tion (HIV gp120 binding to the cell surface receptor CD4). Ala-nine scanning CD4 showed that the HIV gp120-binding siteconsists of several discontinuous segments that could be mod-eled onto a compact region of CD4 (13). Also through alaninescanning, Gibbs and Zoller (14) uncovered specific residuesand regions of a protein kinase likely to be important in cata-lysis, binding MgATP, and binding a peptide substrate. Pro-tein stability can also be probed by alanine scanning asdemonstrated by a series of alanine mutations in the helicalregion of phage T4 lysozyme, which identified three residuesthat have a substantial influence on protein stability (15).These examples demonstrate the power of alanine scanningto correlate different aspects of protein function and stabilitywith structure.

‘‘Inverse alanine scanning’’ offers a labor-intensive alter-native to conventional alanine scanning (16). Each residue ofa parent peptide consisting exclusively of alanines is sepa-rately and sequentially replaced by the 19 nonalanine aminoacids. As usual for scanning mutagenesis, each of these pro-tein variants is separately synthesized and assayed. Inversealanine scanning is perhaps suited only to short peptides.This differs from conventional alanine scanning, which workswell for both large and small proteins.

Although alanine mutagenesis provides a detailed mapof functional epitopes, the method can involve much effort.Each alanine-substituted protein must be separately con-structed, expressed, refolded, characterized, etc. Only thencan the effect of the truncated side chain on protein function-ality be assessed by an in vitro assay of protein activity.


In vivo assays can minimize effort spent on protein purifica-tion and other steps, but such assays are available for onlya subset of interesting proteins. In addition, in vivo assaysdo not offer the control over receptor and ligand concentra-tions possible with in vitro assays, which can be necessaryfor rigorous assessment of binding thermodynamics. Combin-ing alanine scanning with phage display can simplify some ofthe more tedious steps associated with traditional alaninescanning. For example, alanine mutants can be displayed onthe phage surface by fusing the protein under study to aphage coat protein. The displayed proteins can then be ampli-fied in an E. coli host, isolated by phage precipitation, charac-terized by standard DNA sequencing, and examined by an invitro phage-based assay. Alanine scanning proteins displayedon the surface of phage, termed ‘‘turbo alanine scanning,’’ hasprovided insight into the functional epitopes of many proteinsincluding those shown in Table 1.

While turbo alanine scanning expedites the processof functional epitope mapping, combinatorial libraries ofphage-displayed alanine substitutions offers an alternativeto scanning each position individually (Fig. 1). However, toapply a combinatorial library technique, two basic issuesmust be addressed. First, synthesis of the library requires

Table 1 Receptor–Ligand Systems Representative of FunctionalEpitopes Studied by Alanine Scanning Phage-Displayed Proteins(‘‘Turbo Alanine Scanning’’)a

Receptor LigandNo. of mutated

residues Reference

Natriureticpeptide receptor-A

Atrial natriureticpeptide

25 (33)

ErbB3 andErbB4 receptors

Heregulinb egfdomain

45 (34)

Micro-plasminogen

Staphylokinase 9 (35)

IGF bindingprotein 1 and 3

Insulin-likegrowth factor(IGF)

59 (36)

aMutated proteins are shown in italics.


substitution of alanine and wild-type in specific positions overlarge sections of the protein. Second, functional proteins mustbe selected from a library with diversities of up to 1011

alanine-substituted proteins. The use of phage-displayed

Figure 1 Alanine scanning using (a) single mutations where eachamino acid is individually replaced with alanine (three separatemutagenesis reactions required) and (b) combinatorial mutagenesiswhere many amino acids are simultaneously mutated in a singlestep.


protein libraries is appealing because the technique canaddress both issues inherent to combinatorial alaninescanning.

III. COMBINATORIAL SITE-SPECIFICMUTAGENESIS

III.A. Binomial Mutagenesis

As an alternative to conventional alanine-scanning mutagen-esis, several research groups have described methods to accessmultiple alanine substitutions. Oligonucleotide-directed muta-genesis is used for site-specific alanine incorporation intomultiple positions. Binomial substitutions of either alanine oranother amino acid are accessible by conventional oligonucleo-tide synthesis for seven amino acids (labeled with an asteriskin Table 2). For these seven amino acids, changing a singlenucleotide encoding the wild-type amino acid can result in acodon for alanine. For example, the codon GAT encodes theamino acid aspartic acid; replacing A with C in the second posi-tion results in a codon for alanine (GCT). During oligonucleotidesynthesis, addition of a 1:1 ratio of A and C phosphoramidites inthe second position of the codon encodes a 1:1 ratio of asparticacid and alanine in the translated protein library. Librarieswith alanine substitutions in multiple positions can be encodedby degenerate oligonucleotides designed with mutations inmultiple positions.

Mutagenesis with the seven amino acids for which bino-mial mutagenesis is accessible has been used to decipher func-tional epitopes of proteins. Gregoret and Sauer (17) used thetechnique to analyze the effects of multiple alanine substitu-tions on the function and stability of the DNA binding protein,l repressor. Eleven positions of the l repressor helix-turn-helix,a critical region for protein functionality, were mutated to ala-nine or wild-type. Approximately 25% of the alanine-substi-tuted proteins retained activity, which is indicative of therobust information encoding protein folding and function. Ingeneral, the use of combinatorial libraries to study functionalepitopes relies upon the proper folding of alanine-substituted


Table 2 Degenerate Codons Used to Substitute Wild-Type and Another Amino Acid. DNA Degeneracies areRepresented in IUB Code (K¼G=T, M¼A=C, N¼A=C=G=T, R¼A=G, S¼G=C, W¼A=T, Y¼C=T)

Alanine shotgun scanninga Homolog scanningb Proline Scanningc

Wild-type Codon Replacement aa Codon Homolog aa Codon Replacement aa

Ala GST Gly KCT Ser SCA ProArg SST Ala Gly Pro ARG Lys CST ProAsn RMC Ala Asp Thr RAC Asp MMT Pro His ThrAsp� GMT Ala GAM Glu SMT Pro Ala HisCys KST Ala Gly Ser TSC Ser YST Pro Arg SerGlu� GMA Ala GAM Asp SMG Pro Ala GlnGln SMA Ala Glu Pro SAA Glu CMG ProGly� GST Ala GST Ala SST Pro Ala ArgHis SMT Ala Asp Pro MAC Asn CMT Prolle RYT Ala Thr Val RTT Val MYT Pro Leu ThrLeu SYT Ala Pro Val MTC lle CYG ProLys RMA Ala Glu Thr ARG Arg MMG Pro Gln ThrMet RYG Ala Thr Val MTG Leu MYG Pro Leu ThrPhe KYT Ala Ser Val TWC Tyr YYT Pro Leu SerPro� SCA Ala SCA Ala YCT SerSer� KCC Ala KCC Ala YCT ProThr� RCT Ala ASC Ser MCT ProTrp KSG Ala Gly Ser TKG Leu YSG Pro Arg SerTyr KMT Ala Asp Ser TWC Phe YMT Pro His SerVal� GYT Ala RTT lle SYA Pro Ala Leu

aCodons used for alanine shotgun scanning (27). Binomial substitution (wild-type or alanine) is accessible by exchanging a single nucleotide for seven

amino acids (�). Combinatorial alanine scanning of other amino acids requires split-pool synthesis of degenerate oligonucleotides or the listed tetranomial

substitutions.bCodons used in homolog scanning to encode the wild-type residue and a homologous amino acid (29).cCodons that could be used in proline scanning for introducing systematic breaks in protein secondary structure.

448

Avran

tinis

andW

eiss

proteins. Studies of the l repressor described above, T4lysozyme (18), Arc repressor (19), and other systems (reviewedin Ref. 20) have all demonstrated the resilience of protein fold-ing despite extensive alanine substitutions.

To extend binomial mutagenesis beyond the seven aminoacids for which exchange of a single codon nucleotide resultsin a codon encoding alanine, Vernet and coworkers (21,22)reported the split-pool chemical synthesis of degenerate oligo-nucleotides for combinatorial alanine mutagenesis. Split-poolsynthesis of oligonucleotides can be used to couple an alaninecodon onto the nascent oligonucleotide in one reaction vesseland a wild-type codon in a different reaction vessel. After eachtriplet codon is added to the nascent oligonucleotide, the twofractions are pooled. Further splitting, coupling, and poolingcreates combinatorial libraries of oligonucleotides with bino-mial substitution possible for all 20 amino acids. Split-poololigonucleotide synthesis can be used to probe specific positionsof any protein. Altschuh and coworkers (23) used combinatorialalanine scanning with mutagenesis by split-pool synthesizedoligonucleotides to investigate the interface between heavyand light variable domains of an antibody. Combinatorial ala-nine scanning revealed a functional requirement for wild-typeside chains bordering the antigen binding site, which wereascribed to second sphere effects. These results demonstratethe value of combinatorial alanine scanning for identificationof residues contributing indirectly to protein function.Although the split-pool synthesis of oligonucleotides has notbeen used for phage display to date, the technique could bereadily applicable to a phage-display system.

The combinatorial format of alanine-substituted librarieshas also tested the following long debated question relevant toall functional epitope mapping studies. Are the energetic con-tributions of individual amino acids additive? Binomial muta-genesis of l repressor, shotgun scanning (described below),and work by others (24) has demonstrated conclusively that,with a few exceptions, individual side chains contribute bind-ing energy to the receptor–ligand interaction in an essentiallyadditive fashion. In other words, contributions to the DDG fora noncovalent binding interaction largely result from the sum


of energetic contributions by individual side chains. Thisresearch set the stage for large-scale combinatorial alaninemutagenesis techniques, such as shotgun scanning. Shotgunscanning can apply combinatorial alanine mutagenesis to 20or more positions, resulting in exceptionally diverse proteinlibraries, which are rapidly screened using phage-displaytechniques.

III.B. Shotgun Scanning

Shotgun scanning combines the concepts of alanine scanningmutagenesis and binomial mutagenesis with phage-displaytechnology (reviewed in Ref. 25). Libraries of alanine-substi-tuted proteins are displayed on the surfaces of filamentousphage particles for in vitro binding selections (Fig. 2). By dis-playing libraries of alanine-substituted proteins on the surfaceof filamentous phage, successive rounds of selection for a speci-fic binding activity can be used to enrich for side chains contri-buting binding energy to the receptor–ligand interaction(Fig. 3).

Figure 2 Flowchart indicating the steps needed to constructphage-display libraries of alanine-substituted proteins througholigonucleotide-directed mutagenesis.


Phage display simplifies the construction and screeningof combinatorial libraries of alanine-substituted proteins inat least three important ways, which are discussed elsewherein this volume. Briefly, large libraries of proteins (>1010

unique clones) are readily accessible with each unique proteinvariant fused to the surface of a different phage particle. Sec-ondly, after selection for displayed proteins that bind to a tar-get molecule, phage can be amplified in an E. coli host.Additionally, the phage particles encapsulate DNA encodingthe displayed protein; thus, DNA sequencing can be used toidentify selected proteins. The final DNA sequencing step

Figure 3 Screening combinatorial alanine libraries. (a) Librariesof proteins with alanine mutations are constructed using oligonu-cleotide-directed mutagenesis. (b) The phage library is applied toan immobilized receptor and nonbinding phage are washed away.(c) Through successive rounds of binding selection and amplificationin an E. coli host, wild-type side chains are enriched in positionswith energetically favorable contacts.


enables statistical analysis of the frequency of wild-type andalanine in each mutated position.

Shotgun scanning applies degenerate oligonucleotidessynthesized by conventional automated methods. Conven-tional DNA synthesis leads to tetranomial substitution forthe 12 amino acids where single nucleotide exchange for ala-nine substitutions is unavailable (Table 2). A key simplifyingassumption is made during statistical analysis of the selectedand sequenced shotgun scanning clones. Analysis of the ener-getic contribution by specific side chains to receptor–ligandbinding focuses entirely on the distribution of alanine orwild-type in each substituted position. With this simplifica-tion, combinatorial alanine mutagenesis becomes possibleusing standard oligonucleotide synthesis. However, secondaryanalysis of nonwild-type, nonalanine substitutions in eachposition is also possible. For example, statistical analysis toextract information about protein functionality from tetra-nomial mutagenesis has been described by Tidor and cowor-kers (26). Strong selection for nonalanine and nonwild-typeamino acids could reveal unexpected interactions, suchas mutations that confer improved affinity to the ligand. Inaddition, it could be possible to derive DDG values for suchnonalanine mutations.

In the first example of shotgun scanning, the 19 residuesof hGH comprising a large part the high affinity binding sitefor hGHbp were shotgun-scanned in a single library (Fig. 4a)(27). After multiple rounds of selection and amplification inan E. coli host, individual hGHbp binding phage wereidentified by a high throughput phage-based ELISA and sub-jected to DNA sequencing. Focusing entirely upon the distri-bution of alanine and wild-type in each scanned positionrevealed specific positions which were highly conserved asthe wild-type amino acid and other positions with a roughlyeven distribution of alanine and wild-type. By assuming thatthe shotgun scanning selection occurred under equilibriumbinding conditions, estimates of DDG values were derivedfrom the distribution of alanine and wild-type amino acidsin each position following shotgun scanning. Analysis ofhGH in this initial experiment allowed comparison of DDG


Figure 4 Proteins examined by shogun scanning with stick mod-els depicting mutated side chains. (a) Shotgun scanning hGH forbinding to hGHbp, identified seven critical residues (black) outof the 19 residues in contact with the receptor. (From Ref. 27).(b) Streptavidin shotgun scanning revealed side chains essentialfor biotin binding (black), including some residues far from thebiotin-binding site (From Ref. 28). (c) Alanine scanning the heavychain of anti-ErbB2 antibody revealed the side chains that contri-bute to antigen binding (black) (From Ref. 29). (d) Saturationmutagenesis at positions 44 and 53 (black) of GB1 was used tostudy residue pairing in b-sheet stability (From Ref. 30).


values measured by shotgun scanning with DDG values fromconventional alanine scanning mutagenesis. Overall, datafrom hGH shotgun scanning compared favorably with mea-surements by alanine scanning mutagenesis. The shotgunscanning DDG values confirmed that hGH binding to hGHbpis mediated by a compact hot spot of just seven residues out ofthe 19 residues analyzed. Shotgun scanning offered theconvenience of a single round of mutagenesis in multiplepositions with the rapid purification and assay of phage-dis-played alanine-substituted proteins.

Subsequently, several other proteins and receptor–ligandinteractions have been examined by shotgun scanning. Forexample, the femtomolar interaction between the proteinstreptavidin and the small molecule biotin (MW¼ 244 Da)has been dissected by shotgun scanning (Fig. 4b) (28). Theresults demonstrated the importance of previously unreportedhydrophobic residues contributing both direct and indirect con-tacts with biotin. These include residues that are most likelyresponsible for forming the b-barrel structure of streptavidin,residues that interact at the tetramer interface, and residuesthat form a network of extended hydrophobic interactions but-tressing residues in direct contact with biotin.

Sidhu and coworkers (29) used alanine shotgun scanningand a variation called ‘‘homolog shotgun scanning’’ to map theantigen-binding site of an antiErbB2 antibody. Alanineshotgun scanning revealed the antibody side chains thatcontribute to ErbB2 binding (Fig. 4c). These included sol-vent-exposed residues that likely comprise the functional-binding epitope and buried residues that maintain the confor-mation of the residues required for the functional epitope. Aseparate homolog scan, in which library side chains were var-ied as wild-type or a similar amino acid residue, identified asubset of side chains that were intolerant to both alanineand homologous substitutions. The subset of side chains intol-erant to homologous substitutions may be involved in precisecontacts with the antigen. Homolog scanning demonstratesthe potential utility of systematic mutagenesis of proteinsby combinatorial libraries with substitutions different fromalanine. For example, proline scanning could be used for


introducing systematic breaks in protein secondary structure(Table 2).

In an example of using shotgun scanning to examineintrinsic protein stability, as opposed to receptor–ligand mole-cular recognition, Cochran and coworkers (30) used shotgunscanning to examine residue pairing across two strands of theGB1 b-sheet. The GB1 is a small, IgG-binding, model proteinthat has been used to examine individual residue contributionsto b-sheet stability. GB1 shotgun scanning libraries foc-used saturation mutagenesis upon two previously examinedGB1 residues located at cross-strand positions in the b-sheet(Fig. 4d). Properly folded variants were selected from thelibraries through binding to human IgG1 Fc. Selectants fromthe libraries displayed distinct preferences for specific aminoacids at each position. However, the researches were surprisedto find that contributions to b-sheet stability from individualresidues are more important than cross-strand contacts invol-ving specific side chain–side chain interactions. In addition,these experiments demonstrated good agreement betweenshotgun scanning methods with values for protein stabilitymeasured by extensive, conventional experiments. Shotgunscanning can, thus, be viewed as a method for rapidly identify-ing trends in binding activity and=or protein stability, whichcan be followed up by additional experiments.

IV.OTHERAPPROACHES TOPHAGE-DISPLAYEDFUNCTIONAL EPITOPE MAPPING

Residues identified by alanine scanning can be used to guidefurther mutagenesis studies. To investigate the binding inter-face between tissue factor and factor VIIa, Lee and Kelley (31)created libraries in which two tissue factor residues shown byalanine scanning to be critical for high affinity, as well as sur-rounding residues, were randomized to include all aminoacids. Unlike limited randomization throughout a protein,saturated randomization in specific positions allows all possi-ble variants to be covered in a single phage-display library.

Another method to restrict diversity is through ‘‘biased’’phage libraries, in which residues are mutated to a limited


subset of amino acids. Kossiakoff and coworkers (32) used‘‘biased’’ phage-display libraries with targeted residues ofRNase S-peptide constrained as either polar or nonpolar sub-stitutions. Such tailored libraries limited possible interactionspresent during a process of affinity maturation. High-affinityS-peptide variants retained a specific tryptophan residue,which provided a hot spot of binding energy. In general, lim-ited mutagenesis, either by shotgun scanning or other formsof substitution bias, allows larger regions of the protein to beanalyzed simultaneously.

V. CONCLUSION

With molecular recognition being key to essentially all biologi-cal events, methods are needed for rapid, yet detailed, analysisof functional epitopes of proteins. In the past, systematic muta-genesis of many residues in a protein required a tour de forceeffort, yet extensive mutagenesis may be required to map acomplete functional epitope. Combinatorial alanine scanningconnects the expedience of combinatorial libraries with theinsight of site-directed scanning mutagenesis. Phage-displayand combinatorial library techniques simplify these studiesand make it possible to examine large libraries of proteinvariants. These large libraries can provide a more completeunderstanding of the principles of molecular recognition, pro-tein folding, and the relationship between protein structureand function.

REFERENCES

1. Hodges RS, Merrifield RB. Synthetic study of the effect oftyrosine at position 120 of ribonuclease. Int J Pept ProteinRes 1974; 6:397–405.

2. Fersht AR, Shi JP, Knill-Jones J, Lowe DM, Wilkinson AJ,Blow DM, Brick P, Carter P, Waye MMY, Winter G. Hydrogenbonding and biological specificity analyzed by protein engi-neering. Nature (London) 1985; 314:235–238.


3. Jespers L, Jenne S, Lasters I, Collen D. Epitope mapping bynegative selection of randomized antigen libraries displayedon filamentous phage. J Mol Biol 1997; 269:704–718.

4. Rossenu S, Dewitte D, Vandekerckhove J, Ampe C. A phagedisplay technique for a fast, sensitive, and systematic investi-gation of protein–protein interactions. J Protein Chem 1997;16:499–503.

5. Cain SA, Ratcliffe CF, Williams DM, Harris V, Monk PN.Analysis of receptor=ligand interactions using whole-moleculerandomly mutated ligand libraries. J Immunol Methods 2000;245:139–145.

6. Myers RM, Lerman LS, Maniatis T. A general method forsaturation mutagenesis of cloned DNA fragments. Science(Washington, DC) 1985; 229:242–247.

7. Chen G, Dubrawsky I, Mendez P, Georgiou G, Iverson BL. Invitro scanning saturation mutagenesis of all the specificitydetermining residues in an antibody binding site. ProteinEng 1999; 12:349–356.

8. Cunningham BC, Wells JA. Comparison of a structural and afunctional epitope. J Mol Biol 1993; 234:554–563.

9. Clackson T, Wells JA. A hot spot of binding energy in ahormone–receptor interface. Science (Washington, DC) 1995;267:383–386.

10. Clackson T, Ultsch MH, Wells JA, de Vos AM. Structural andfunctional analysis of the 1:1 growth hormone–receptor com-plex reveals the molecular basis for receptor affinity. J MolBiol 1998; 277:1111–1128.

11. Pons J, Rajpal A, Kirsch JF. Energetic analysis of anantigen=antibody interface: alanine scanning mutagenesisand double mutant cycles on the HyHEL-10=lysozyme interac-tion. Protein Sci 1999; 8:958–968.

12. Arkin MR, Wells JA. Probing the importance of second sphereresidues in an esterolytic antibody by phage display. J Mol Biol1998; 284:1083–1094.

13. Ashkenazi A, Presta LG, Marsters SA, Camerato TR, RosenthalKA, Fendly BM, Capon DJ. Mapping the CD4 binding site for


human immunodeficiency virus by alanine-scanning mutagen-esis. Proc Natl Acad Sci USA 1990; 87:7150–7154.

14. Gibbs CS, Zoller MJ. Identification of electrostatic interactionsthat determine the phosphorylation site specificity ofthe cAMP-dependent protein kinase. Biochemistry 1991; 30:5329–5334.

15. Blaber M, Baase WA, Gassner N, Matthews BW. Alaninescanning mutagenesis of the a-helix 115–123 of phage T4 lyso-zyme: effects on structure, stability and the binding of solvent.J Mol Biol 1995; 246:317–330.

16. Vetter SW, Keng Y-F, Lawrence DS, Zhang Z-Y. Assessment ofprotein-tyrosine phosphatase 1B substrate specificity using‘‘inverse alanine scanning’’. J Biol Chem 2000; 275:2265–2268.

17. Gregoret LM, Sauer RT. Additivity of mutant effects assessedby binomial mutagenesis. Proc Natl Acad Sci USA 1993; 90:4246–4250.

18. Heinz DW, Baase WA, Matthews BW. Folding and function ofa T4 lysozyme containing 10 consecutive alanines illustratethe redundancy of information in an amino acid sequence. ProcNatl Acad Sci USA 1992; 89:3751–3755.

19. Brown BM, Sauer RT. Tolerance of Arc repressor to multiple-alanine substitutions. Proc Natl Acad Sci USA 1999; 96:1983–1988.

20. Bowie JU, Reidhaar-Olson JF, Lim WA, Sauer RT. Decipher-ing the message in protein sequences: tolerance to amino acidsubstitutions. Science (Washington, DC) 1990; 247:1306–1310.

21. Chatellier J, Mazza A, Brousseau R, Vernet T. Codon-basedcombinatorial alanine scanning site-directed mutagenesis:design, implementation, and polymerase chain reactionscreening. Anal Biochem 1995; 229:282–290.

22. Chatellier J, Vernet T. Combinatorial scanning site-directedmutagenesis. Curr Innovations Mol Biol 1997; 4:117–132.

23. Chatellier J, Van Regenmortel MH, Vernet T, Altschuh D.Functional mapping of conserved residues located at the VLand VH domain interface of a Fab. J Mol Biol 1996; 264:1–6.


24. Wells JA. Additivity of mutational effects in proteins.Biochemistry 1990; 29:8509–8517.

25. Morrison KL, Weiss GA. Combinatorial alanine-scanning.Curr Opin Chem Biol 2001; 5:302–307.

26. Hu JC, Newell NE, Tidor B, Sauer RT. Probing the rolesof residues at the e and g positions of the GCN4 leucinezipper by combinatorial mutagenesis. Protein Sci 1993; 2:1072–1084.

27. Weiss GA, Watanabe CK, Zhong A, Goddard A, Sidhu SS.Rapid mapping of protein functional epitopes by combina-torial alanine scanning. Proc Natl Acad Sci USA 2000; 97:8950–8954.

28. Avrantinis SK, Stafford RL, Tian X, Weiss GA. Dissecting thestreptavidin–biotin interaction by phage-displayed shotgunscanning. Chembiochem 2002; 3:1229–1234.


30. Distefano MD, Zhong A, Cochran AG. Quantifying b-sheetstability by phage display. J Mol Biol. 2002; 322:179–188.

31. Lee GF, Kelley RF. A novel soluble tissue factor variant withan altered factor VIIa binding interface. J Biol Chem 1998;273:4149–4154.

32. Dwyer JJ, Dwyer MA, Kossiakoff AA. High affinity RNaseS-peptide variants obtained by phage display have a novel ‘‘hot-spot’’ of binding energy. Biochemistry 2001; 40:13491–13500.

33. Li B, Tom JYK, Oare D, Yen R, Fairbrother WJ, Wells JA,Cunningham BC. Minimization of a polypeptide hormone.Science (Washington, DC) 1995; 270:1657–1660.

34. Jones JT, Ballinger MD, Pisacane PI, Lofgren JA, Fitzpatrick VD,Fairbrother WJ, Wells JA, Sliwkowski MX. Binding interaction ofthe heregulin b egf domain with ErbB3 and ErbB4 receptorsassessed by alanine scanning mutagenesis. J Biol Chem 1998;273:11667–11674.


35. Jespers L, Van Herzeele N, Lijnen HR, Van Hoef B, De Maeyer M,CollenD, Lasters I. Arginine 719 inhuman plasminogen mediatesformation of the staphylokinase:plasmin activator complex.Biochemistry 1998; 37:6380–6386.

36. Dubaquie Y, Lowman HB. Total alanine-scanning mutagenesisof insulin-like growth factor I (IGF-I) identifies differentialbinding epitopes for IGFBP-1 and IGFBP-3. Biochemistry1999; 38:6386–6396.


13

Selections for Enzymatic Catalysts

JULIAN BERTSCHINGER, CHRISTIAN HEINIS,and DARIO NERI

Institute of Pharmaceutical Sciences,Swiss Federal Institute of Technology,

Zurich, Switzerland

I. INTRODUCTION

Enzymes are extremely powerful catalysts and their profi-ciency and diversity typically exceeds the performance ofman-made chemical catalysts that are commonly used inindustrial chemistry. An impressive example for the stunningefficiency of enzymes is provided by orotidine 50-phosphatedecarboxylase. The uncatalyzed reaction has a half-life of 78million years, whereas the half-life of the catalyzed reactionis only 18msec. This corresponds to a rate acceleration(kcat=kuncat) of 1.4� 1017 (1).

The use of the catalytic power of enzymes for the synth-esis of commodity products and fine chemicals could be an

461

attractive, cost-effective, and environmentally friendly alter-native to chemical catalysts. Enzymes may also find medicalapplications in areas such as prodrug activation and treat-ment of metabolic disease (2,3).

However, naturally occurring enzymes usually do notmeet the requirements for industrial or medical applications.For the synthesis of chemicals, it is desirable to use high sub-strate concentrations, which may lead to inhibition of the bio-catalyst (4). In addition, the enzyme should be stable, accept acertain range of substrates whilst producing defined mole-cules without side products. For therapeutic applications,the enzyme should be nonimmunogenic, exhibit high specifi-city for the target molecule, and not interfere with other phy-siologic processes in the patient. Thus, the engineering oftailor-made enzymes, with improved properties, representsan important goal of modern protein engineering.

There are two general avenues to engineer enzymes. Thefirst is rational design, which requires information aboutstructure, catalytic mechanisms, and molecular modeling todesign an enzyme de novo or to alter an existing one. Thealternative to design is evolution: a process, which involvesseveral cycles of protein diversity generation, followed byselection or screening of mutants with desired characteristics.The procedure is analogous to a Darwinian process, whereonly the fittest survive. Applying Darwinian principles inthe test tube to form desired phenotypes is called in vitro evo-lution. In vitro evolution has been successfully applied in thelaboratory to routinely produce antibodies with high affinityto their target molecule (5), enzymes with novel activities(6,7), ribozymes (8), and even allosteric ribozymes (9). Thisbook chapter will focus on the in vitro evolution of enzymesusing phage display.

I.A. In Vitro Evolution of Enzymes

In vitro evolution of proteins consists of three basic steps.First of all, sequence diversity must be generated. In a secondstep, the created mutant phenotypes (i.e., the function per-formed by the protein) have to be linked to the corresponding

462 Bertschinger et al.

genotypes (i.e., the genetic information encoding the protein).And finally, the mutants with the desired biological activitymust be selected from the ensemble of proteins. In the caseof enzymes, the biological activity is their ability to acceleratechemical reactions. Thus, enzymes could be selected by virtueof their ability to form product molecules. This requires thelinkage of the product molecules to the genotype, which isresponsible for their formation.

A wide range of approaches has been developed to engi-neer proteins with catalytic activity. Various in vitro (10)and in vivo procedures, such as auxotrophic complementation(11) or, in the case of catalytic antibodies, immunization withtransition-state analogues (TSAs) (12) and reactive immuni-zation with a suicide inhibitor (SI) (13), demonstrated thatcatalysts with improved or novel activities could be gener-ated. In vitro selection systems are potentially applicable ina much broader way compared to in vivo procedures, becausethey could allow the selection step to take place under non-physiologic conditions such as elevated temperatures, highor low salt concentrations, extreme pH values or even inorganic media. In addition, due to their complexity, in vivosystems may find ways to pass the selection barrier of theexperiment by a number of different strategies, which maylead to the evolution of undesired phenotypes.

I.B. Linking Genotype and Phenotype byPhage Display

There are two main strategies to link a genotype to thecorresponding protein phenotype. The first possibility is tolink the protein physically to the DNA encoding it. Thisway of linkage is fundamentally different from what naturedoes. In nature, genes, the proteins they encode and the pro-ducts of their activity are held together in a cell. Therefore,the linkage between phenotype and genotype is achieved bycompartmentalization. Most procedures applied in proteinengineering use the physical linkage between genotype andphenotype. Phage display is one example for this. The pro-tein to be engineered is displayed on the phage coat by

Selections for Enzymatic Catalysts 463

fusing it to a phage coat protein (pIII, pVI, or pVIII). Severalenzymes could already be displayed on phage (14–17).

Phage display does not take place entirely in vitro, asphage need to be assembled and amplified in living bacteria,but the selection step does. Phage are relatively resistant toa variety of environments, which makes them appropriatefor use in enzyme engineering.

II. SELECTION METHODS

A variety of methods for the selection of catalysts using phagedisplay has been developed. The different approaches can bedivided in two groups: selection of catalysts by binding anddirect selections for catalytic activity.

II.A. Selecting Catalysts by Binding

Whenever possible, selection of phage–enzymes by means oftheir binding properties may be a practical methodology, asit may allow to physically isolate phage–enzymes with desiredcharacteristics from the phage pool. For the selection of cata-lysts, affinity-purification on TSAs or mechanism basedsuicide inhibitors has been considered.

II.A.1. Selections by Binding to Transition-StateAnalogues

The transition state theory relates the rate of a reaction to thedifference in Gibbs energy between the transition state andthe ground state (Fig. 1). The higher this difference is, theslower the reaction rate. By binding the transition state,enzymes can lower its energy level and accelerate the reaction.

The dissociation constant of the complex between catalystand a reaction’s transition state is defined asKTS. Efficient cata-lysts bind the transition state very tightly, and therefore KTS isvery small. KTS is an important parameter to characterize theproficiency of an enzyme. Using transition state theory and therelations of the Gibbs free energy to equilibrium constants, KTS

can be described by the parameters kcat, kuncat and KS. kcatdenotes the rate constant for reaction of the enzyme–substrate


complex,KS its dissociation constant, andkuncat the rate constantfor the uncatalyzed reaction of substrate to product.

KS

kcat=kuncat¼ KTS ð1Þ

Eq. (1) shows the relation between the three parameters KS,kcat, kuncat, and the dissociation constant KTS for theenzyme-transition state complex. Enzymes can bind the tran-sition state very tightly. Orotidine 50-phosphate decarboxy-lase that was mentioned at the beginning of this chapterhas a KM of 7� 10�7M and a kcat over kuncat ratio of1.4� 1017 (1). Using Eq. (1) and assuming that KS�KM, itcan be calculated that this remarkable enzyme binds thetransition state with a dissociation constant of 5� 10�24M.

Many attempts have been made to create catalytic anti-bodies, which, by binding and stabilizing the transition state,

Figure 1 Free energy diagram for catalysis. The dissociation con-stant KTS correlates quantitatively with DDG�. S stands for sub-strate, P for product, ES and EP for the enzyme–substrate=productcomplexes. TS� and ETS� denote the free and the enzyme-boundtransition state, respectively.


accelerate a reaction. Molecules mimicking the transitionstate (TSAs) were synthesized. With an ideal TSAs perfectlymimicking the actual transition state of the reaction, the rateacceleration of the selected catalysts would match the differ-ential affinity for the TSA vs. the substrate [Eq. (1)].

In most cases, catalytic antibodies were raised by immu-nization against a TSA. This has led to catalytic antibodies fora number of reactions. Rate accelerations are limited to abouta factor of 102–104, occasionally 106 (18). As a strategy toimprove the activity of the antibodies, affinity selectionsagainst a TSA with antibody fragments displayed on phagewere performed to optimize the affinity for TSAs (19). In thisstudy, the humanized antibody 17E8 with esterase activitywas affinity-matured against a phosphonate TSA. The selec-tions resulted in antibody variants with two- to eightfoldhigher affinity for the phosphonate TSA. Surprisingly, noneof the selected mutants showed improved catalytic activitycompared to the parent antibody. By contrast, a weaker bind-ing variant was identified to have a twofold higher catalyticactivity. In a further study (20), residues remote from theactive site of 17E8 were mutated. Panning of the libraryyielded a mutant with 10-fold improved catalytic efficiency(kcat=KM). However, higher affinity to the TSA did not resultin a parallel increase in catalytic activity. The increase of rateacceleration observed was due to a lower substrate affinity(20). This is not surprising because the selection pressure isfor TSA binding and not for catalysis. Even if TSA bindingwas equivalent to catalytic activity, it would be very difficultto obtain binders with very low KTSA. Rate enhancements byenzymes range from 106 to 1017 with an average around 1010

(21). When assuming that KM� 1mM, KTSA would have to bein the order of 10�13M to obtain a catalytic antibody withactivity comparable to enzymes. However, the selection ofantibodies by phage display with affinities as high as10�13M remains a formidable challenge. Notably, one of theantibody–hapten complexes with the lowest dissociation con-stant (48 fM) was isolated by yeast display (5). In addition,a TSA is never a perfect mimic of a true transition stateand their synthesis can be very difficult. Therefore, to select


catalytic antibodies against TSAs using phage display or anyother method is of limited scope. It may be useful for findingcatalysts without counterpart in nature. Better results maybe achieved if, after selecting for binding, the library isscreened (22) or selected for catalysis (23).

II.A.2. Selections by Binding to Suicide Inhibitors

Mechanism based suicide inhibitors were used as an alterna-tive to TSAs (24,25). Enzymes tolerate a suicide inhibitor assubstrate but the catalytic cycle does not proceed until itsend. Instead, the reaction is trapped in an intermediate,where the suicide inhibitor is covalently bound to the activesite of the enzyme. When affinity tags are coupled to the sui-cide inhibitor, enzymes labeled with the affinity tag throughthe suicide inhibitor can then be selected from repertoires ofmutants using affinity-purification.

Using suicide inhibitors for selection by binding, thespecificity of subtilisin 309 could be changed (24). Subtilisinshave broad substrate specificity but exhibit a preference forhydrophobic residues and very low reactivity toward chargedresidues in P4 position. A wild-type subtilisin called Savinasefrom B. lentus was displayed on phage fused to the pIII coatprotein. The phage–enzyme particles were successfullypanned on streptavidin-coated beads after labeling with bioti-nylated suicide inhibitors bearing the hydrophobic residuealanine in the P4 position.

In order to change the specificity of Savinase, librarieswere created where residues 104 and 107 forming part of theS4 binding pocket were randomized. For the selections biotiny-lated suicide inhibitorswere usedwith lysine instead of alaninein the P4 position. Savinase variants were isolated after threerounds of panning with a more than 100-fold enhanced activityfor the substrate with lysine in P4 position compared to thewild-type enzyme. The isolated clone did not have any detect-able activity for the substrate with alanine in the P4 position.

Mechanism based inhibitors may be superior to TSAsbecause more of the catalytic cycle is being selected for. Insome cases, the combined use of TSA and suicide inhibitors


may yield biocatalysts of exceptional quality (26). For exam-ple, the immunization of mice with a hapten combining anelement mimicking the transition state of the reaction (a tet-rahedral sulphone) with a diketone mechanism-based inhibi-tor has yielded antibodies with exceptional aldolase activities(rate accelerations > 108 and efficient turnover). In principle,it should be possible to apply a similar strategy using phagedisplay instead of immunization.

Similar to selections by binding to TSAs, selections bybinding to suicide inhibitors have their limitations. First ofall, it is impossible to select for product release and turnoverwhen using suicide inhibitors. These steps can be rate limitingin catalysis (27). Secondly, in reactions where the TSA or thetrapped suicide inhibitor resembles the product, the selectedcatalysts may suffer from product inhibition and low turnover.In conclusion, selecting catalysts by binding (using phage dis-play or any other method) is probably only effective wherechanging the substrate selectivity and not enhancement ofrate acceleration or turnover is of main importance.

II.B. Selections for Catalytic Activity

In nature, enzymes are selected directly (but not solely) fortheir catalytic activity. When looking at the amazing profi-ciency of natural enzymes, direct selection for catalysis seemsto be the most effective way of evolving catalysts. Proteinswith catalytic activity must have several abilities.

� The catalyst must be able to bind the substrate effi-ciently at given concentrations. This property is char-acterized by the Michaelis–Menten constant KM. KM

values of natural enzymes vary from 1mM to 1nMand so do the effective, physiological concentrationsof the corresponding substrates (28).

� Specificity is also important. The catalyst should beable to catalyze the conversion of one particular sub-strate in the presence of others.

� The catalyst must convert the substrate bound in theactive site to product with a higher rate than that of thefree substrate in solution (rate acceleration, kcat=kuncat).


� The catalyst should be specific for the formation of oneparticular product (or few products). In solution, freesubstrate might turn into several, different products,whereas the enzymatic reaction only yields one pro-duct (e.g. one stereoisomer vs. a racemic mixture).

� And finally, the product should readily dissociate fromthe active site to yield free enzyme, which is ready tobind new substrate (turnover, kcat). Sometimes, thecatalyst has to change its conformation before it is ableto bind substrate again. In order to select for turnover,the substrate should be in excess over the enzyme andselection pressure should be directed to the conversionof all or most of the substrate molecules.

In order to obtain man-made catalysts of proficiency com-parable to the one of natural enzymes, selection pressure mustbe applied on all properties mentioned above simultaneously.

When phage display is used to engineer enzymes, thereaction product must either be immobilized on the enzyme–phage particle to allow the isolation of catalytic proteins usingan antiproduct affinity reagent, or the enzyme–phage parti-cles must be selectively eluted from a solid support upon cata-lysis. Several procedures have been developed to selectdirectly for catalytic activity. They can be divided in two mainclasses: intramolecular, single-turnover selections and inter-molecular, multiple-turnover selections.

II.B.1. Intramolecular, Single-Turnover Selections

An approach in which the reaction substrate was covalentlyattached to the pIII minor coat protein of filamentous phage,displaying a nuclease, was first published in 1998 (29). Intra-molecular conversion of the substrate into product was meantto provide the basis for the selection of active catalysts from alibrary of mutants. Upon generation of the reaction product,the enzyme–phage particle would be released from a solidsupport (Fig. 2).

In order to immobilize the substrate of the nucleaseon the phage coat nearby the enzyme, a helper phage wasgenerated displaying an acidic amphiphilic peptide at the


N-terminus of the pIII coat protein. The acidic peptide andthe enzyme were codisplayed on phage after infection ofE. coli (containing a phagemid) with the modified helper phage.The acidic peptide formed a heterodimeric coiled-coil complexwith a basic amphiphilic peptide. By coupling the substrate ofthe reaction to the basic peptide, the substrate could be nonco-valently attached to the phage adjacent to the enzyme. Inorder to achieve a covalent linkage between the two peptides,one cysteine was introduced at the C-terminus of each peptideand therefore, upon removal of reducing agents, a covalent dis-ulfide bridge between the two peptides was formed. In thisway, the substrate could be coupled to the phage coat sitespecifically in vicinity to the enzyme (Fig. 3).

The nuclease used [staphylococcal nuclease (SNase)]preferentially hydrolyzes the phosphodiester bonds of single-stranded RNA, single-stranded DNA, and duplex DNA at A,U- or A, T-rich regions upon activation by Ca2þ. The substratefor the enzymatic reaction, a biotinylated oligodeoxynucleotidewas chemically coupled to the basic peptide. After addition ofsubstrate-basic peptide conjugate to the phage–enzyme parti-cles followed by removal of reducing agents the enzyme–phageparticles could be immobilized on streptavidin coated beads viathe biotin–streptavidin interaction. The cleavage reaction wasinitiated by addition of Ca2þ. By virtue of the catalytic activityof SNase, the enzyme–phage particles were released from the

Figure 2 Upon intramolecular catalysis the enzyme–phage parti-cle is released from a solid support.


solid support. Phage displaying SNase were shown to bereleased 100 times more efficiently than a control antibodyphage. It appeared that a small fraction of the phage leakedoff the support during the assay.

The methodology described above could in principle beapplied to any other reaction involving substrate cleavage.However, the enzyme must be maintained in an inactive stateduring immobilization of the enzyme–substrate–phage parti-cles on the solid support. Whenever reversible inactivationof the enzyme is not possible, capture of active enzymes mustbe performed with a product-specific reagent. In the case ofbimolecular condensation reactions, in which bond formationresults in phage immobilization on solid support, the regula-tion of enzymatic activity is not necessary.

Figure 3 Immobilization of the substrate on phage with a disul-fide bridged, heterodimeric coiled–coil complex. The acidic peptidewas codisplayed with the enzyme on phage.


A very similar approach as the one with SNase (29) wasused for the selection of metalloenzymes by catalytic elution(30). The method should be applicable to phage displayingenzymes whose activity depends on presence of a cofactor,provided that the apoenzyme is still capable of binding to itssubstrate. In this method, a metalloenzyme, which requireda metal ion cofactor, was displayed on phage. After complexa-tion of the metal ion by a chelator, the inactivated enzyme–phage particles were immobilized on a solid-support coatedwith substrate. After addition of the metal ion cofactor, activeenzymes transformed the substrate into product. Because theenzyme exhibited only low affinity for the product, theenzyme–phage particles were eluted from the solid support(Fig. 4).

In this study (30), metallo-b-lactamase BCII (bLII) fromBacillus cereus was used as model enzyme. The enzyme’sstructure contains two zinc ions. One zinc ion, coordinatedby three histidine residues, is essential for catalytic activity.Thus, phage bound enzymes could be inactivated and reacti-vated by the incubation with EDTA followed by addition ofzinc sulfate. In model experiments, phage displaying bLIIwere enriched over irrelevant phage displaying an inactiveTEM-1 b-lactamase mutant (25) by a factor of up to 185. Alibrary of 5� 106 mutants was then generated by error-pronePCR. The mean activity of the library was 1.6% of the activityof the wild-type. After two rounds of selections, 11 clones wererandomly picked, monoclonal enzyme–phage particles wereprepared, and their activities were measured. The specificityconstants kcat=KM, ranged from 20% to 170% of the wild-typevalue. Further on, the thermostability of the selected cloneswas characterized. The selected variants had two to fourmutations, all of which were distant from the active site. Allthe mutants were at least 6 times less stable than the wild-type enzyme on phage.

The method described above allowed the isolation ofactive mutants from a repertoire of enzymes. However, theactivity of the best mutant was not significantly higher thanthat of the wild-type enzyme. Better results may be achievedby choosing another strategy to generate the library (e.g.


Figure 4 Selection of metalloenzymes by catalytic elution. Cofac-tor depleted, inactivated enzyme–phage particles were adsorbed ona solid support. Upon catalysis, the enzyme dissociated from theproduct and was eluted.


mutating defined regions of the enzyme), or by constructing alarger library.

In another selection method, the substrate of a reactionwas fused to the enzyme itself (31), instead of being immobilizedonapIII coatproteinneighboring theonedisplaying theenzyme(29). Subtiligase, a double mutant of subtilisin BPN0 (32) thatcatalyzes the ligation of peptides was displayed on phage. Toobtain an optimized subtiligase active site, 25 active site resi-dues were randomly mutated in groups of four or five, yieldingsix different libraries with more than 109 individual memberseach. Mutants were selected by their ability to ligate a biotiny-lated peptide onto their own extended N-terminus. The func-tional enzyme–phage particles were isolated by capture onimmobilized neutravidin. Prior to selection experiments, itwas demonstrated that ligation was occurring intramolecu-larly. Bymixing active subtiligase phage lacking anN-terminalextension with a phage displaying catalytically inactive subtili-gase containing theN-terminal extension, itwas shown that thebiotinylated peptide was not ligated onto any phage.

After five to seven rounds of selection with each library,several clones were assayed for their catalytic activity and themost active were subcloned into parent subtiligase that didnot contain the N-terminal extension. Some of the selectedmutants had increased ligase activity (about twofold). How-ever, the most frequently isolated mutant showed a lowerligase activity than subtiligase but display was improved bya factor 10. Other mutations yielded mutants with better oxi-dative resistance. These results reflect that the selectionmethod not only favored subtiligase variants with improvedcatalytic activity but also mutants with other propertiesimproving functional enzyme display.

In a further experimental strategy, proximity couplingwas used to retain the product of the enzymatic reactionlinked to the enzyme–phage particle responsible for it (33).The strategy could be classified as intramolecular, single-turnover or intermolecular, multiple-turnover selection. Itinvolves two chemically independent reactions. A catalyticreaction that leads to the product and a chemical cross-linking reaction in which substrate (and eventually product)


is linked to the phage. There are two avenues to a phagelabeled with tagged product. Firstly, the substrate reacts withthe phage coat and then is converted into product by theenzyme. Secondly, the substrate is converted into productbefore the cross-linking reaction takes place. The first casewould correspond to an intramolecular, single-turnover selec-tion, whereas the latter would be consistent with an intermolecular, multiple-turnover selection. The order of the reac-tions will depend on the relative rates of the two reactions.

In this study, it was assumed that cis reactions would befavored over reactions in trans by proximity effects (33),thereby allowing the labeling of only those phage responsiblefor product formation. For the cross-linking reaction, malei-mides were chosen because they are known to react withthiols, and in alkaline solutions, with amino groups. To testthe approach, the Klenow and Stoffel fragments of DNA poly-merase I from Escherichia coli and Thermus aquaticus weredisplayed on phage as fusion to pIII. An oligonucleotide pri-mer with a maleimidyl group at its 50 end was used as sub-strate. The product was tagged by addition of biotinylateddUTP to the 30 end of the primer by the catalytic action ofthe polymerase (Fig. 5).

The display of the polymerases was difficult. It wasestimated that only one in a thousand phage particles dis-played a fusion protein (33). In selection experiments, phagedisplaying Stoffel fragments could be enriched by a factor26 over Klenow phage when they were incubated at 60�C dur-ing the reaction. This reflects the higher thermostability ofthe Taq-polymerase derived fragment compared to the Kle-now fragment. Further selection experiments with Stoffelphage with different catalytic activities were performed.Enrichment factors of up to 123 in favor of the more activeStoffel fragment were achieved. Thus, proximity couplingcould be used as a general strategy for the selection of cataly-tic activities from large repertoires. However, there are somepoints to be discussed.

As a first problem, products could diffuse to anotherphage which displays an inactive enzyme and react with thephage coat proteins. This would lead to the isolation of false


positives and reduce the efficacy of the selection system. Thehigher the proficiency of the active phage enzymes, the moreproduct molecules would probably be found on inactiveenzyme–phage particles. In the work described above (33),only a small fraction of phage displayed polymerase. Phageexclusively carrying helper phage-derived pIII were treatedwith trypsin, so as to render them incapable of mediatinginfection (34). Therefore, very few catalysts displayed onphage were surrounded by a large pool of noninfective phage,which possibly acted as a sink for the products diffusing awayfrom the active enzyme–phage particles. The probability for adiffusing product being captured by an inactive phage enzymewould be low. In this way, the noninfective phage could pre-vent the isolation of infective phage displaying inactiveenzymes. However, the recovery of the active enzyme–phageparticles could be more efficient if all product molecules wereretained on the phage responsible for their formation.

It was suggested for bimolecular condensation reactionsto immobilize the reactive substrate before addition of the

Figure 5 Selection of catalytically active DNA polymerase byphage display and proximity coupling. By the extension of the sub-strate primer with biotinylated dUTP in a template dependent man-ner, the product of the reaction was tagged for selection. Themaleimidyl group reacted with the phage coat or the enzyme,thereby linking the product of the phenotype to the correspondinggenotype.


tagging compound (33). But prelabeling of the enzyme–phageparticles could possibly prevent most substrate moleculesfrom reaching the active site of the enzyme displayed on pIIIdue to the linker length or sterical hindrance. A further pos-sibility to prevent the labeling of inactive enzyme–phageparticles with product could be the use of compartmentaliza-tion strategies (35).

Targeted immobilization of the substrate on phage, asdescribed above for the peptides forming a heterodimericcoiled–coil complex (29),was also achieved by inserting calmodu-lin between pIII and the enzyme (36). Calmodulin was function-ally displayed on phage as pIII-calmodulin-enzyme fusion withdifferent enzymes as fusion partners. Calmodulin binds verytightly to the peptide CAAARWKKAFIAVSAANRFKKIS in acalcium dependent manner (37). By fusing the substrate to thecalmodulin-binding peptide, the substrate was kept in vicinityto the enzyme displayed on phage. Catalytically activeenzyme–phage particles were able to catalyze the formation ofproduct.Becauseof the linkagebetweenproductandcalmodulin,itwaspossible to isolatephagedisplayingactiveenzymeusinganaffinity reagent against the product. As the interaction betweenpeptide and calmodulin is calcium dependent, captured phagewere eluted by the addition of a calcium chelator (Fig. 6).

Model selections were performed with three enzymes:submut (mutant of subtilisin from B. subtilis) (36), biotinligase (BirA) from E. coli (38), and the rat endopeptidase tryp-sin (His57Ala mutant) (38). BirA catalyzes the biotinylation ofa lysine residue in a 13-mer peptide. The H57A mutant oftrypsin cleaves the sequence GGHR=DYKDE, whereassubmut catalyzes the hydrolysis of AAHY=DYKDE. To adaptthe reaction substrates to the selection scheme describedabove, the substrate peptides for the three reactions werefused to the calmodulin-binding peptide. Phage displayingsubmut were enriched over phage displaying an irrelevantenzyme by a factor 54 (36). In model selection experiments,phage displaying either BirA or trypsin H57A were isolateddue to their catalytic activity. In the case of BirA, the phagetiters recovered from the selection were 4–800 times lowerwhen either the biotin–acceptor peptide or the biotin was


omitted. When substrate peptides containing the wrong clea-vage site were added to the calmodulin-tagged phage display-ing the H57A mutant of trypsin, the phage titers recoveredfrom the selection experiment were 15–2000 times lower com-pared to the one with the correct substrate peptide (38).

Since the results from the model selections were promis-ing, a library with more than 5� 105 trypsin mutants wasconstructed. The aim was to isolate a H57A trypsin variantwith identical substrate specificity but improved catalyticactivity. After three to four rounds of panning, 30 trypsin

Figure 6 Calmodulin-tagged phage enzyme for the selection ofenzymatic activity. The substrate was noncovalently tethered tothe enzyme–phage particle. Upon catalysis, the reaction productwas used as an affinity tag for the isolation of the correspondingphage. The calmodulin=peptide complex was dissociated by additionof a calcium chelator.


mutants were screened for their endoproteolytic activity.None was found to show an endopeptidase activity similarto H57A trypsin. H57A trypsin was also present in thelibrary, but was not recovered. This result suggests that theselection conditions applied were probably not stringentenough. Because the substrate peptide is tethered adjacentto the enzyme, the effective substrate concentration wasabout 10�3M. In order to cleave one peptide at this concentra-tion during a time period that starts with the addition ofsubstrate peptides and ends with the wash during affinity-purification of the phage, only very moderate catalyticactivity is required.

Intramolecular, single-turnover selections have twoweaknesses. Firstly, an enzyme–phage particle is onlyenriched over nonactive counterparts if the enzyme can accel-erate product formation over the rate of the uncatalyzed reac-tion. The tight attachment of substrate near the enzyme leadsto high effective substrate concentrations. If it is assumedthat one substrate molecule is present in a sphere with theenzyme as its center and that the diameter of this sphere is10nm, the effective concentration of the substrate can thencalculated to be approximately 3mM. Thus, it is difficult toselect for lower KM values. Irrespective of whether an enzymemutant had either a KM of 10�4 or 10�7M, the enzyme wouldbe saturated with substrate in both cases and, therefore, theformation of product would occur with comparable rates.

Secondly, selection pressure is for rate acceleration, butnot for product release and turnover. However, high turnovernumber is the prerequisite for an efficient catalyst, becauseone catalyst should be able to catalyze the formation of manyproduct molecules. In intramolecular, single-turnover selec-tions, it only is possible to select for higher turnover by short-ening the reaction time. For a reaction with a kcat of 1 sec�1,the half-life is 0.7 sec. This value is reduced to 0.007 sec for akcat of 100 sec�1. In practice, such differences in reaction timeare difficult, if not impossible, to resolve.

On the other hand, the two disadvantages mentionedabove are also the strength of intramolecular, single-turnoverreactions. The high effective substrate concentration and the


requirement for the formation of only one product moleculeleads to the possibility that proteins with very modest cataly-tic activities could be selected. This could make the intra-molecular, single-turnover approach especially useful whenevolving catalysts de novo.

II.B.2. Intermolecular, Multiple-TurnoverSelections

In contrast to the selection methods described so far, inter-molecular, multiple-turnover selection strategies shouldallow the selection pressure for all parameters and steps ofthe catalytic cycle, which are important to be optimized toobtain an efficient enzyme. These parameters and steps are:substrate–enzyme interaction (KM), rate acceleration(kcat=kuncat), product release, and turnover (kcat).

In order to select for product release and turnover, theenzyme to be selected should be forced to process a high num-ber of substrate molecules within a given reaction time. Forexample, the selection can be based on the number of productmolecules. The more product molecules are formed, the morelikely it should be for an enzyme–phage particle to be selected.With a kcat of 100 sec�1, an enzyme would yield 100 timesmore product molecules than another enzyme with a kcat ofonly 1 sec�1, therefore increasing significantly the probabilityfor the enzyme–phage particle to pass the selection barrier.

As already mentioned above, the work of Jestin et al. (33)can be seen either as an intramolecular or intermolecularapproach, depending on which of the two reactions—cross-link-ing or primer elongation—is faster. To our knowledge, only oneother studyhasbeenpublishedso far,whichuses intermolecular,multiple-turnover selections combined with phage display (23).

In this study (23), catalytic antibodies were raised byimmunization against a hapten mimicking the transitionstate of glycosidic bond cleavage. Approximately 100 clonesthat bound the TSA were generated by established hybridomatechnology (39). The heavy and the light chains of theseantibodies were shuffled to yield a library of about 104 indivi-dual phage-Fab fragments. A substrate molecule containing a


difluoromethylphenol moiety was designed. Upon glycosidicaction of the displayed Fab fragment on this substrate, thedifluoromethylphenol moiety generated a reactive quinonemethide species at or near the active site. Any neighboringnucleophile can be alkylated by this reactive intermediate,thereby potentially attaching the substrate molecule rightafter cleavage of the glycosidic bond covalently to the phage-Fab complex. By coupling the substrate to a solid support orto biotin prior to enzymatic treatment, phage immobilizedor covalently labeled with the reaction product were selected.

In order to test this trapping strategy, substrate containingthedifluoromethylphenolmoietynext to the glycosidic bond to becleaved was biotinylated. The substrate was incubated with b-galactosidase andwas found to act as substrate-alkylating agentof this enzyme. Biotinylated b-galactosidase could be trapped onstreptavidin coatedELISAplates. Although the enzymewas bio-tinylated, it retained its catalytic activity. This nondestructivetrapping of the enzyme is important because it may allow select-ing clones with the ability to undergo multiple turnovers.

After four rounds of selections with the phage displayingthe library of heavy and light chains, the best Fab fragmentexhibited a rate acceleration (kcat=kuncat) of 7�104 for the hydro-lysis of p-nitrophenyl-b-galactopyranoside (kcat¼ 0.0001 sec�1,and KM ¼ 0.53mM) (Fig. 7). This Fab fragment was comparedwith antibodies obtained from simple hybridoma screening.Twenty-two monoclonal antibodies were assayed for their cata-lytic activity and the best hydrolyzed p-nitrophenyl-b-galacto-pyranoside with a rate acceleration of only 102. Therefore, byshuffling the heavy and light chains of only 100 antibodies, anincrease of catalytic activity by 700-fold was achieved. Althoughmechanism based inhibitors and suicide substrates have beendescribed for a number of reactions, it will be difficult to gener-ally convert this information into trapping compounds, which donot disturb the performance of the catalyst after their covalentlinkage near the active site. The design of such sophisticatedtrapping substrates is even more complicated for reactionswhere only little is known about their mechanism.

The authors did not investigate whether the Fab dis-played on phage were able to cleave several gylcosidic bonds.


From the experiments performed with soluble b-galactosi-dase, it can be assumed that several turnovers should be fea-sible. However, it is possible that the active sites of the Fabfragments become more and more blocked with increasingnumber of cleaved glycosidic bonds. Thus, it remains to beseen whether this approach will be useful for the selectionof enzymes exhibiting high turnover numbers.

III. DISCUSSION

III.A. Selection for Catalysis: Comparisonof Phage Display with Other Methods

A number of systems to link a protein phenotype to the corre-sponding genotype have been reported. The linkage has been

Figure 7 Procedure for the mechanism-based immobilization ofantibody-phage particles that catalyzed the hydrolysis of a galacto-pyranoside substrate. (Adapted from Ref. 23.)


achieved by several methods such as phage display (14), pep-tides displayed as lac repressor fusions (‘‘peptides on plas-mids’’) (40), and cell–surface display systems (6). All of thesemethods include a step of transformation of living cells andtherefore, the library sizes are limited to 109–1010 individualmembers. However, larger libraries are important to obtainaccess to a maximum of different mutants. Therefore, it wouldbe desirable to perform directed evolution with libraries aslarge as possible. In order to increase library size, a numberof in vitro selection systems based on cell-free transcriptionand translation have been developed. The in vitro translatedpolypeptide was coupled to its encoding mRNA in a ribosomecomplex (41), or through a puromycin derivative (42,43). Inaddition, in vitro expressed proteins were fused to theirencoding DNA through different strategies (44–46). Invitro compartmentalization was used as a third class ofgenotype–phenotype linkage (35,47). Of all these methods tolink genotype and phenotype, only few have been used toselect for catalysis: cell–surface display (6), phage display(14), and in vitro compartmentalization (35,47).

Phage display has some advantages and disadvantagescompared to other protein selection methods. An importantstrength of phage display in selections for catalysis is the rela-tive robustness of the phage particles under a variety ofconditions. It may be necessary to evolve a catalyst at lowor high pH values, high substrate concentrations which aretoxic for living cells, in high salt solutions, or at moderatelyelevated temperatures. In every one of these situations,cell–surface display would not be the system of choice becausethe cells would most probably die during the selection proce-dure. Similarly, due to the sensitivity of RNA to degradation,RNA-peptide fusions or ribosome display cannot be an alter-native to phage display. A compartmentalized system asproposed by Tawfik and Griffiths (35) is not well suited eitherbecause the conditions for the enzymatic reaction must becompatible with the requirements for efficient in vitro tran-scription and translation. In order to evolve enzymes undernonphysiologic conditions, the expression of phenotype andcoupling to its genotype must be separated temporally and


spatially from the selection procedure. This is an importantrequirement for all in vitro systems. Phage display takesplace partially in vivo and in vitro and therefore, fulfills thiscriterion. An additional advantage of phage display is theease with which enzyme–phage particles can be purified. Thisis an important practical aspect because it may be necessaryto remove excess of substrate or product molecules, which arenot immobilized on the enzyme–phage particle before startingthe capture of the phage tagged with product.

Covalent DNA-protein fusions are a very interesting alter-native tophagedisplay.After compartmentalized invitro expres-sion they could be purified and then be used in the selection stepunder a large variety of conditions because of the robustness ofDNA. In addition, through DNA-protein fusions produced invitro, much larger sequence space becomes accessible than withphage display because larger libraries can be created.

One of the biggest problems of phage is the inefficiency ofprotein display on the phage surface (33,36,38). By the use ofphage vectors instead of phagemids, display was improved.Better display correlated with increased efficiency of selectionprotocols (38). Several attempts have been undertaken toimprove display of proteins on phage. Mutants of pVIII coatproteins have been isolated that increased display on pVIII(48), a helper phage was developed which does not providewild-type pIII after superinfection (49), and signal peptideswere selected from a library to enhance display of a DNA-polymerase fragment on phage (50).

II.B. Possible Improved Selection Systems Basedon Phage Display

Intermolecular, multiple-turnover selections without usingphage display have been reported. These selection systemstake place fully (11) or partially in vivo (47), or in vitro (35).The use of compartments to retain the product molecules invicinity to the genotype responsible for their formation is com-mon to all three systems. The compartments used are cells (11),cells in combination with reversed micelles (47), or reversedmicelles (35). It might be possible to develop a selection systemwhere enzyme–phage particles together with substrate are


enclosed in a reversed micelle, which acts as a cell-like com-partment. Depending on the substrate concentration, theactivity of the enzyme variant in the compartment, the timegiven to the enzyme for the formation of product, and the sizeof the compartment (which influences the effective concentra-tion of the enzyme in the reversed micelle), different amountsof substrate will be converted into product. If the productmole-cules could then be immobilized on the phage at a given timepoint before the water-in-oil emulsion is broken, enzyme–phage particles labeled with product molecules could beselected by affinity-purification on a product-binding ligand.By applying stringent washes during affinity-purification,phage labeled with many product molecules on their surfacewould be favored due to avidity effects. Such a selection schememay allow high selection pressure for all the parameters whichneed to be optimized to obtain an efficient catalyst.

II.C. Implications for Biotechnology andDrug Discovery

Enzymes are becoming more and more important in biotech-nology and therapeutic applications. In chemical and pharma-ceutical industries, enzymes and microorganisms are used forthe manufacture of a number of fine chemicals and drugs. In1990, among the top 50 best selling pharmaceuticals, 10 wereproduced from low-molecular-weight fermentation (51).Industrially significant processes catalyzed by enzymes arestarch hydrolysis and production of b-lactam antibiotics(51). In medicine, enzymes are used in diverse fields includingantibody dependent prodrug therapy (ADEPT), therapy ofmetabolic diseases (2,3), and leukemia treatment (52).

However, as already mentioned at the beginning of thischapter, enzymes often do not meet all the requirements forthe applications for which they are intended to be used.Therefore, technologies must be developed to engineertailor-made biocatalysts. Several attempts have been madeto achieve this goal, of which in vivo systems based on auxo-trophic complementation (11) and reactive immunizations(26) were the most successful. But the former of these systems


are limited to special cases where product formation iscoupled to survival, and the latter to catalytic antibodies,which exhibit only moderate rate accelerations. So far, in con-trast to the engineering of antibodies, directed evolutionusing phage display or any other method has not led to profi-cient enzymes which are used in medicine or biotechnology.This may be due to the lack of powerful and general selectionmethods. Furthermore, the choice of the residues to bemutated is crucial for the success of the selection experiment.It is difficult to predict which of the systems described in thischapter will be important in the future, but we believe that anefficient selection procedure should be feasible in vitro andshould make use of compartmentalization to retain reactionproducts close to the genotype encoding the enzyme (intermo-lecular, multiple-turnover selections). Strategies must befound for the targeted immobilization of products on the gen-otype at any desirable time point. In addition, protocols mustbe developed for the improved display of proteins.

We anticipate that the development of efficient in vitroselection strategies (based on phage display or not) will makeit possible to isolate improved enzymes for a variety of appli-cations. In medicine, for example, human nonimmunogenicproteases could be used as molecular knives to cleavecell–surface receptors or other proteins important for thesurvival of tumor cells. In industry, engineered enzymesmay help to synthesize fine chemicals with less side productsand lower energy input.

REFERENCES

1. Radzicka A, Wolfenden R. A proficient enzyme. Science 1995;267:90–93.

2. Bhatia J, Sharma SK, Chester KA, Pedley RB, Boden RW,Read DA, Boxer GM, Michael NP, Begent RH. Catalytic activ-ity of an in vivo tumor targeted anti-CEA scFv::carboxypepti-dase G2 fusion protein. Int J Cancer 2000; 85(4):571–577.

3. Russell CS, Clarke LA. Recombinant proteins for geneticdisease. Clin Genet 1999; 55(6):389–394.


4. Buchholz K, Kasche V. Reaktoren und prozesstechnik. In:Buchholz K, Kasche V, eds. Biokatalysatoren und Enzymtech-nologie. Weinheim, New york, Basel, Cambridge, Tokyo: VCHVerlagsgesellschaft, 1997:273–279.

5. Border ET, Midelfort KS, Wittrup KD. Directed evolution ofantibody fragments with monovalent femtomolar antigen-binding affinity. Proc Natl Acad Sci USA 2000; 97:2029–2034.

6. Olsen MJ, Stephens D, Griffiths D, Daugherty P, Georgiou G,Iverson BL. Function-based isolation of novel enzymes from alarge library. Nat Biotechnol 2000; 18:1071–1074.

7. Santoro SW, Schultz PG. Directed evolution of the site specifi-city of Cre recombinase. Proc Natl Acad Sci USA 2002;99:4185–4190.

8. Chapman KB, Szostak JW. In vitro selection of catalytic RNAs.Curr Opin Struct Biol 1994; 4:618–622.

9. Breaker RR. Engineered allosteric ribozymes as biosensorcomponents. Curr Opin Biotechnol 2002; 13(1):31–39.

10. Griffiths AD, Tawfik DS. Man-made enzymes—from design toin vitro compartmentalization. Curr Opin Biotechnol 2000;11:338–353.

11. Altamirano MM, Blackburn JM, Aguayo C, Fersht AR. Direc-ted evolution of new catalytic activity using the alpha=beta-barrel scaffold. Nature 2000; 403:617–622.

12. Patten PA, Gray NS, Yang PL, Marks CB, Wedemayer GJ,Boniface JJ, Stevens RC, Schultz PG. The immunologicalevolution of catalysis. Science 1996; 271:1086–1091.

13. Barbas CF, Heine A, Zhong G, Hoffmann T, Gramatikowa S,Bjornestedt R, List B, Anderson J, Stura EA, Wilson IA,Lerner RA. Immune versus natural selection: antibodyaldolases with enzymatic rates but broader scope. Science1997; 278:2085–2092.

14. Smith GP. Filamentous fusion phage: novel expression vectorsthat display cloned antigens on the virion surface. Science1985; 228:1315–1317.

15. HuftonSE,Moerkerk PT,MeulemannsEV, deBruineA,ArendsJW, Hoogenboom HR. Phage display of cDNA repertoires: the


pVIdisplay systemand its applications for the selectionof immu-nogenic ligands. J Immunol Methods 1999; 231:39–51.

16. Malik P, Terry TD, Gowda LR, Langara A, Petukhov SA,Symmons MF, Welsh LC, Marvin DA, Perham RN. Role ofcapsid structure and membrane protein processing in deter-mining the size and copy number of peptides displayed onthe major coat protein of filamentous bacteriophage. J Mol Biol1996; 260:9–21.

17. Sidhu SS, Weiss AG, Wells JA. High copy display of largeproteins on phage for functional selections. J Mol Biol 2000;296:487–495.

18. Reymond JL. Catalytic antibodies for organic synthesis. In:Fessner WD, ed. Topics in Current Chemistry. Berlin, Heidel-berg: Springer Verlag, 1999:59–93.

19. Baca M, Scanlan TS, Stephenson RC, Wells JA. Phage displayof a catalytic antibody to optimize affinity for transition-state analog binding. Proc Natl Acad Sci USA 1997;94:10063–10068.

20. Arkin MR, Wells JA. Probing the importance of second sphereresidues in an esterolytic antibody by phage display. J Mol Biol1998; 284:1083–1094.

21. Fersht A. Chemical catalysis. In: Fersht A, ed. Structure andMechanism in Protein Science. New York: W.H. Freemanand Company, 1999:60.

22. Tawfik DS, Green BS, Chap R, Sela M, Eshhar Z. catELISA: afacile general route to catalytic antibodies. Proc Natl Acad SciUSA 1993; 90:373–377.

23. Janda KD, Lo LC, Lo CHL, Sim MM, Wang R, Wong CH,Lerner RA. Chemical selection for catalysis in combinatorialantibody libraries. Science 1997; 275:945–948.

24. Legendre D, Laraki N, Graslund T, Bjornvard ME, Bouchet M,Nygren P, Borchert TV, Fastrez J. Display of active subtilisin309 on phage: analysis of parameters influencing the selectionof subtilisin variants with changed substrate specificity fromlibraries using phosphonylating inhibitors. J Mol Biol 2000;296:87–102.


25. Soumillion P, Jespers L, Bouchet M, Marchand-Brynaert J,Winter G, Fastrez J. Selection of beta-lactamase on filamen-tous bacteriophage by catalysis activity. J Mol Biol 1994;237:415–422.

26. Zhong GF, Lerner RA, Barbas CF. Broadening the aldolasecatalytic antibody repertoire by combining reactive immuniza-tion and transtition state theory: new enantio- and diastereos-electivities. Angew Chem Int Ed 1999; 38:3738–3741.

27. Fersht A. Measurement and magnitude of individual rate con-stants. In: Fersht A, ed. Structure and Mechanism in ProteinScience. New York: W.H. Freeman and Company, 1999:164–168.

28. Fersht A. Enzyme-substrate complementarity and the use ofbinding energy in catalysis. In: Fersht A, ed. Structure andMechanism in Protein Science. New York: W.H. Freemanand Company, 1999:364–368.

29. Pedersen H, Holder S, Sutherlin DP, Schwitter U, King DS,Schultz PG.Amethod for directed evolution and functional clon-ing of enzymes. Proc Natl Acad Sci USA 1998; 95:10523–10528.

30. Ponsard I, Galleni M, Soumillion P, Fastrez J. Selection ofmetalloenzymes by catalytic activity using phage display andcatalytic elution. Chem Biochem 2001; 2:253–259.

31. Atwell S, Wells JA. Selection for improved subtiligases byphage display. Proc Natl Acad Sci USA 1999; 96:9497–9502.

32. Abrahmsen L, Tom J, Burnier J, Butcher KA, Kossiakoff A,Wells JA. Engineering subtilisin and its substrates for efficientligation of peptide bonds in aequeous solution. Biochemistry1991; 30:4151–4159.

33. Jestin JL, Kristensen P, Winter G. A method for the selectionof catalytic activity using phage display and proximity cou-pling. Angew Chem Ind Ed 1999; 38:1124–1127.

34. Kristensen P, Winter G. Proteolytic selection for proteinfolding using filamentous bacteriophages. Fold Des 1998;3:321–328.

35. Tawfik DS, Griffiths AD. Man-made cell-like compartments formolecular evolution. Nat Biotechnol 1998; 16:652–656.


36. Demartis S, Huber A, Viti F, Lozzi L, Giovannoni L, Neri P,Winter G, Neri D. A strategy for the isolation of catalytic activ-ities from repertoires of enzymes displayed on phage. J MolBiol 1999; 286:617–633.

37. Montigiani S, Neri G, Neri P, Neri D. Alanine substitutions incalmodulin-binging peptides result in unexpected affinityenhancement. J Mol Biol 1996; 258:6–13.

38. Heinis C, Huber A, Demartis S, Bertschinger J, Melkko S,Lozzi L, Neri P, Neri D. Selection of catalytically active biotinligase and trypsin mutants by phage display. Protein Eng2001; 14:1043–1052.

39. Kohler G, Milstein C. Continuous cultures of fused cellssecreting antibody of predefined specificity. Nature 1975;256:495–497.

40. Cull MG, Miller JF, Schatz PJ. Screening for receptor ligandsusing large libraries of peptides linked to the C terminus of thelac repressor. Proc Natl Acad Sci USA 1992; 89:1865–1869.

41. Hanes J, Schaffitzel C, Knappik A, Pluckthun A. Picomolaraffinity antibodies from a fully synthetic native libraryselected and evolved by ribosome display. Nat Biotechnol2000; 18:1287–1292.

42. Nemoto N, Miyamato-Sato E, Husimi Y, Yanagawa H. In vitrovirus: bonding of mRNA bearing puromycin at the 30-terminalend to the C-terminal end of its encoded protein on the ribo-some in vitro. FEBS Lett 1997; 414:405–408.

43. Wilson DS, Keefe AD, Szostak JW. The use of mRNA displayto select high-affinity protein-binding peptides. Proc Natl AcadSci USA 2001; 98:3750–3755.

44. Doi N, Yanagawa H. STABLE: protein-DNA fusion system forscreening of combinatorial protein libraries in vitro. FEBSLett 1999; 457:227–230.

45. FitzGerald K. In vitro display technologies—new tools for drugdiscovery. Drug Discov Today 2000; 5:253–258.

46. Kurz M, Gu K, Al-Gawari A, Lohse PA. cDNA-protein fusions:covalent protein-gene conjugates for the in vitro selection ofpeptides and proteins. Chem Biochem 2001; 2:666–672.


47. Ghadessy FJ, Ong JL, Holliger P. Directed evolution of poly-merase function by compartmentalized self-replication. ProcNatl Acad Sci USA 2001; 98:4552–4557.

48. Sidhu SS, Weiss GA, Wells JA. High copy display of large pro-teins on phage for functional selections. J Mol Biol 2000;296:487–495.

49. Rondot S, Koch J, Breitling F, Dubel S. A helper phage toimprove single-chain antibody presentation in phage display.Nat Biotechnol 2001; 19:75–78.

50. Jestin JL, Volioti G, Winter G. Improving the displayof proteins on filamentous phage. Res Microbiol 2001; 152:187–191.

51. The application of biocatalysis to the manufacture of fine che-micals. In: Roberts SM, Turner NJ, Willetts AJ, Turner MK,eds. Introduction to Biocatalysis Using Enzymes and Microor-ganisms. Cambridge, UK: Cambridge University Press,1995:140–187.

52. Ertel IJ, Nesbit ME, Hammond D, Weiner J, Sather H. Effec-tive dose of L-asparaginase for induction of remission in pre-viously treated children with acute lymphocytic leukemia: areport from Childrens Cancer Study Group. Cancer Res1979; 39:3893–3896.


14

Antibody Humanization and AffinityMaturation Using Phage Display

JONATHAN S. MARVIN andHENRY B. LOWMAN

Department of Antibody Engineering, Genentech,Inc., South San Francisco, California, U.S.A.

I. INTRODUCTION

Three significant developments in molecular biology havebeen central in promoting the growing role of antibodies astools for biochemical and biological research and as a signifi-cant class of molecules for drug development. First, followingthe establishment of reliable methods for the generation ofhigh-affinity, high-specificity monoclonal antibodies from thenatural immune response of mice and other species (1),recombinant DNA techniques have made it possible to isolatethe corresponding complementary DNA (cDNA) encodingthese antibodies, determine their sequences, and with the

493

help of structural information, optimize their antigen-bindingproperties through manipulation of their genes. Second, com-binatorial approaches to protein engineering such as phagedisplay (2) have made it possible to explore many more var-iants of antibodies than would typically be accessible in asite-directed mutagenesis approach (discussed in earlierchapters). Third, the elucidation of high-resolution molecularstructures of antibodies have revealed some of the generalfeatures required for structural stability and antigen recogni-tion (3,4). These approaches provide not only a means for therapid optimization of antigen-binding properties of existingantibodies as discussed in this chapter, but also a means forthe discovery of novel antibody specificities from immunized,nonimmunized (naıve), or synthetic sources of diversity, asdiscussed in later chapters. Humanization of nonhuman anti-bodies and affinity maturation of antibodies from hybridomaor diversity-library sources are two areas of antibody optimi-zation that can often be most efficiently addressed usingantibody–phage display.

In general, the display of antibodies on phage is similarto that described for other proteins. Polyvalent display maybe useful for the identification of antigen-binding antibodiesfrom naıve antibody–phage libraries (5). However, monova-lent display may offer advantages by reducing avidity effectsduring antigen-binding selections (6,7). Two preferred for-mats have been used for the display of antibody variabledomains on phage, often in the form of a phagemid vector(7–9): (1) Fab-phage in which the variable (VH) and first con-stant (CH1) domains of the heavy chain are fused to a portionof the gene III protein (gIIIp) of filamentous phage, while thevariable and constant (VL and CL) domains of the light chainare expressed as a soluble chain from the same vector, result-ing in expression of a complete antigen-binding fragment(Fab) and (2) scFv-phage in which the VL and VH domainsare linked and fused to a portion of the gIIIp as a single poly-peptide, resulting in expression of a ‘‘single-chain’’ variablefragment (scFv). An advantage of the former method maybe to preserve a monomeric form of the antibody formore efficient binding selections (discussed below), while an

494 Marvin and Lowman

advantage of the latter can be the formation of diabodies (10)which can improve the avidity (apparent affinity) underappropriate conditions, leading to more efficient recovery oflow-affinity binders (6,7).

In the first part of this chapter, we describe the applica-tion of phage display to humanization, often a key step in theprocess of converting monoclonal antibodies from nonhumanspecies to pharmaceutical drug candidates. The need forhumanization was highlighted by early clinical experienceswith murine and chimeric (i.e., containing murine variableand human constant domains) antibodies. Immunogenicresponses in humans to antibodies originating in other spe-cies has the potential to render a therapeutic antibody ineffec-tive after initial dosing, and worse, may give rise topotentially fatal hypersensitivity reactions (anaphylaxis).Transplanting the entire variable domains to an antibodyscaffold of human constant domains, producing a chimericantibody (see Fig. 1) does not remove the potential for antifra-mework antibodies. For example, when administered tohuman, the murine antibody OKT3 gave rise to human anti-murine antibodies, some of which were directed to the vari-able region (11), which is also of murine origin in chimericantibodies (Fig. 1). A rat antihuman antibody directed tothe CAMPATH-1 antigen also proved to be immunogenic inhumans, and this was ultimately overcome through humani-zation (12,13). Although transgenic mice producing humanimmunoglobulins are also a source of monoclonal antibodiesfor therapeutic development, humanization remains animportant antibody engineering endeavor because of theremarkable and differing specificities of antibodies derivedfrom various sources. Such differences may translate intodramatic differences in biological activity for molecules beingconsidered for drug development. Indeed, antibodies derivedby different means may often differ in their fine specificityfor epitopes on a given antigen and hence in their biologicaleffects such as blocking an antigen from binding receptor orcrosslinking a cell-surface receptor to induce agonist activity;see for example Refs. 14,15. Humanization by phage displayrepresents a special case of in vitro affinity maturation in

Antibody Humanization and Affinity Maturation Using Phage Display 495

which only selected, often nonantigen-contacting residues arechanged in order to optimize activity, while presenting theessential antigen-binding epitope (or paratope) in the contextof human framework and constant regions. Humanized anti-bodies derived from parental murine hybridomas such asanti-HER2 (16), anti-IgE (17), anti-CD11a (18), and anti-VEGF in both IgG form (19) and an affinity-matured Fab(20) are demonstrating success in clinical trials and as mar-keted therapeutics.

Figure 1 Strategies for using phage display in the humanizationand affinity maturation of antibodies. Murine-derived antibodyregions are shown as shaded domains; human antibody domainsare shown as open domains. Murine residues selected from FRlibraries are depicted as gray dots, and optimized mutations selectedfromCDR libraries are depicted as black dots. The six CDRs are indi-cated as six lines within the variable domains (VH and VL) of theantibody.


In the second part of this chapter, we discuss affinitymaturation, which together with humanization can be seenas part of a unified process of antibody optimization (Fig. 1).Antibodies derived from human, nonhuman, or syntheticsources may benefit from optimization for antigen bindingthrough substitutions at positions within and outside theset of antigen-contacting positions. This is often termed ‘‘invitro affinity maturation’’ by analogy to the process of naturalaffinity maturation, including recombination and somatichypermutation, during maturation of the adaptive immuneresponse in vertebrates. As applied to therapeutic antibodies,the motivations for this type of antibody engineering mayinclude the need to improve potency and efficacy. However,relevant to therapeutic as well as diagnostic, reagent, orindustrial uses of antibodies, motivations for affinity matura-tion may also include considerations of cost-of-goods, manu-facturing capacity, assay sensitivity, or catalytic efficiency.In the second part of this chapter, we consider approachesto affinity maturation for human, humanized, or nonhumanantibodies.

II. HUMANIZATION USING PHAGE DISPLAY

The humanization of antibodies derived from mouse or otherspecies is an optimization process by which the essential anti-gen-binding characteristics of a parental monoclonal antibody(often a murine IgG) are transferred to a human ‘‘scaffold’’antibody—that is, an immunoglobulin with constant (CL,CH1–CH2–CH3) and variable (VL, VH) framework regions(FRs) derived from human genes (Fig. 1). The process typi-cally involves replacing the six complementarity-determiningregions (CDRs) of the human antibody with those from theparental antibody. Despite the homology of human and non-human (e.g., mouse or rat) framework regions, this direct‘‘CDR-swap’’ does not always result in a functional antibodywith the same antigen-binding affinity observed for the par-ental (21). Structural studies of antibodies suggested that afew key residues within the FRs of the antibody variable


domains can greatly influence the conformation of the CDRloops in a given scaffold. Indeed, appropriate substitution ofparental residues for human residues at these positions canrestore antigen binding and activity (3,12,22–24). The huma-nization of antibodies through structure-based, site-specificmutagenesis has been described and reviewed elsewhere(25). These efforts often require the investigation of a rela-tively small number of point mutations; however, since theeffects of combinations of mutations in the framework is notalways additive, many more variants may need to be testedin order to produce a humanized antibody with similar activ-ity to the parental one. Phage display provides a tool for thefacile generation of diversity libraries easily large enough toencompass many of the framework positions that have beenfound key to achieving humanization. Humanized antibodiescan then be selected for binding to antigen from librariesrandomized to contain the parental (nonhuman) or humanresidue at these positions (Fig. 1).

II.A. Choice of Human Scaffold

Several approaches to the selection of scaffolds for humaniza-tion have been described. In general, these can be grouped asfollows: (1) selection of a common variable-region scaffold, forexample, VH subgroup III, VL kappa subgroup I for CDRgrafting and subsequent FR mutations (26,27); (2) selectionof an antibody-specific, human variable FR region with closesequence similarity to the original nonhuman antibody, fol-lowed by specific FR mutations (28); and (3) construction ofvariable-region libraries (representing the diversity of humangermline VH and VL families) with additional syntheticdiversity at key framework positions (29).

Other approaches have abandoned the selection of a coreframework region in favor of simply ‘‘resurfacing’’ the paren-tal antibody with residues found at the surfaces of humanantibodies, the idea being that solvent-exposed residues arelikely to dominate immunogenic reactions (30). Another inter-esting approach involves purely a domain-shuffling strategyin which the VH domain of a parental antibody is first


randomly recombined with a human VL library and selectedfor antigen binding; thereafter, the resulting human VLdomain is similarly shuffled with a human VH domain library(31). While clearly successful in some cases, this approachmay not be generally efficient at preservation of the parentalantibody epitope (29), a clear motivation for humanization asopposed to discovery of novel antibodies.

Here, we focus on the common-scaffold approach tohumanization, with randomization at a small set of frame-work residues to confer the necessary structural determi-nants for presentation of the CDRs to provide antigenrecognition. This approach offers several advantages in thedevelopment of therapeutic antibodies (16). First, the heavy-chain VH subgroup II and light-chain VL kappa subgroup Ifamilies are abundantly represented in the human repertoire.To be sure, antihuman antibodies have been found inhumans, and even antihuman-CDR antibodies are known;however, immunogenicity may be minimized by selectingthe most abundantly represented germline families as huma-nization scaffolds. Clinical experience with anti-HER2, anti-IgE, anti-CD11a, and anti-VEGF antibodies has proven thatthey are indeed low in immunogenicity, supporting the useof this scaffold. The use of a single scaffold also facilitateshumanization of new antibodies because new CDRs can bemodeled onto the same starting scaffold. Finally, proteinexpression, formulation, and clinical experience with a givenscaffold in the context of multiple indications provide a grow-ing database for improving the development of therapeuticantibodies with similar chemical and physical properties.

II.B. Design of Humanization Libraries

Because humanization can typically be achieved by exchan-ging the parental framework residues for residues found inthe human scaffold, the diversity of framework librariesdesigned for humanization can in principal be limited to two-fold diversity at each randomized position (see Table 1)—encoding the parental residue or the human residue (27). Onthe other hand, slightly more diverse randomization can yieldalternative FR residues that also restore antigen binding (27).


Antibody humanization using phage display was reportedfor the murine antihuman VEGF antibody A4.6.1 using thecommon-scaffold approach (27), and the methodologydescribed for this antibody illustrates the basic steps in huma-nization using phage display. As the starting point for phage-displayed libraries, the CDRs of the murine antibody wereinserted in place of the CDRs of human consensus frameworkto produce the variant hu2.0 (Fig. 2). The Fab form of this anti-body was displayed monovalently on phage using a phagemidconstruct in which the heavy-chain VH–CH1 domains were

Table 1 A Typical Set of Limited-Diversity Codons forPhage-Displayed Antibody Framework Region (FR) Libraries inAntibody Humanization

Chain Consensus FR Codon Degeneracy Residuesa

VL M4 MTG ATG, CTG M, LVL F71 TYC TTC, TAC F, YVH A24 RYC GCT, ACT,

GTC, ATCA, T, V, I

VH V37 RTC GTC, ATC V, IVH F67 NYC ATC, GTC,

CTC, TTC,ACC, GCC,CCC, TCC

I, V, L, F, T, A, P, S

VH I69 WTC ATC, TTC I, FVH R71 CKC CGC, CTC R, LVH D73 RMC GAC, ACC,

AAC, GCCD, T, N, A

VH K75 RMG AAG, GCG,ACG, GAG

K, A, T, E

VH N76 ARC AAC, AGC N, SVH L78 SYG CTG, GCG,

GTG, CCGL, A, V, P

VH A93 DYG ATG, GTG,TTG, TCG,ACT, GCG

M, A, V, L, S, T

VH R94 ARG AAG, AGG R, K

a Note: The illustrated randomization has 1.6� 106 diversity, and is similar to thatdescribed in the humanization of murine A4.6.1; however, other schemes are possi-ble (27). The murine FR residues found in A.4.6.1 are underlined.


linked to the C-terminal domain of gIIIp, and the light-chainVL–CL domains were expressed as a soluble protein from thesame phagemid. Compared to the murine–human chimericform of A4.6.1, the CDR swap humanization variant was atleast 4000-fold reduced in binding affinity to VEGF (27).Following VEGF-binding selections with a library selectivelyrandomized at 13 framework positions, a humanized variant(called hu2.10) was selected for binding to VEGF with affinityabout sixfold weaker than that of the chimeric A4.6.1 antibody.Compared to the original CDR-swap construct, hu2.10 con-tained selected substitutions (including two that were neitherof human nor murine origin) at only eight framework posi-tions: VL 71 and VH 37, 71, 73, 75, 76, 78, and 94; see Fig. 2.

II.C. Further Optimization of HumanizedAntibodies

While a framework library approach was successful in gener-ating >400-fold improvement in antigen-binding affinity ascompared with a simple CDR-swap (27), higher-affinity var-iants of humanized anti-VEGF A.4.6.1 were identifiedthrough molecular modeling and additional point mutations(19). In particular, the humanized antibody Fab-12 displayedantigen-binding affinity within twofold of that of the chimericA.4.6.1. This antibody, in the form of a full-length humanizedIgG1, is known as Avastin2 (see Fig. 2), and is being studiedfor the treatment of solid tumors (32).

Ultimately, affinity maturation approaches as describedbelow can be employed to obtain the highest affinity for anti-gen at a given epitope. As discussed below, these approaches,as well as structure-based design, have been applied tofurther optimization of the humanized A4.6.1 anti-VEGFantibody (see Fig. 2).

III. IN VITRO AFFINITY MATURATIONOF ANTIBODIES

The affinity, or strength of binding, of any protein for itsligand is, by definition, the equilibrium ratio of the unbound


Figure 2 Sequences of the variable regions of the human kappalight-chain subgroup I, heavy-chain subgroup III consensus and fourversions of A.4.6.1, an antihuman VEGF antibody. The firstsequence is that of the murine antibody A.4.6.1. The secondsequence corresponds to a human consensus sequence for VH sub-group III, Vk subgroup I (50). The third sequence, hu2.0 representsthe CDR-swap version used as the starting point for phage-displayedlibraries. The fourth sequence, hu2.10, corresponds to the finalphage-derived sequence. The fifth sequence, Fab-12 (19) correspondsto the final humanized antibody (known in IgG form as Avastin2),including changes made after phage humanization. The sixthsequence corresponds to affinity-matured Fab-12 (20) and is also


(Caption continued from previous page) known as Lucentis2. Theseventh sequence corresponds to an on-rate enhanced version(59). The numbering system is that of Kabat et al. (50) and is shownto indicate the boundaries of the framework regions. Complemen-tarity-determining regions as defined by a combination of sequencehypervariability and structural data (see text for details) are shownin brackets. Mutations in the sequence of each humanized version ascompared to the preceding sequence are underlined. Frameworkresidues that have frequently been changed during humanizationof other antibodies on this scaffold (16–19) are double-underlinedin the human consensus sequence.


antibody and ligand to that of the complex. The thermody-namic equilibrium constant can be defined by the relationshipof the concentrations of the components of the binding reac-tion measured at equilibrium (see Refs. 33,34) or by measur-ing association and dissociation rate constants prior toequilibrium using methods such as surface plasmon reso-nance (SPR) (35). The association of the monomericantigen-binding fragment (Ab) of the antibody with antigen(Ag) can be described by the chemical Eq. (1):

Abþ Ag 9 Ab � Ag ð1Þ

For the interaction of an antibody Fab fragment with itscognate antigen, we can calculate the equilibrium (dissocia-tion) constant, Kd, as a function of the Fab, antigen, andFab–antigen complex concentrations at equilibrium ([Ab],[Ag], and [Ab�Ag], respectively), according to Eq. (2):

Kd ¼ ½Ab�½Ag�½Ab � Ag� ð2Þ

When the component concentrations are expressed inmolar units (nM, pM, etc.), it is apparent that Kd has unitsof concentration, and that Kd decreases (approaching zero)as the antibody–antigen interaction becomes tighter (i.e.,more complex and less free component at equilibrium).

The kinetic approach also permits calculation of the equi-librium dissociation constant via statistical mechanics (seeRef. 35) using Eq. (3):

Kd ¼ kd=ka ð3Þ

Here, ka represents the association rate constant forformation of the antigen–antibody complex (often expressedin units of M�1 sec�1 for a two-component reaction), and kdrepresents the rate constant for dissociation of the complex(often expressed in units of sec�1). Hence, Kd again has unitsof concentration, and decreases as the antibody–antigeninteraction becomes tighter (i.e., faster association or slowerdissociation).


Any measurement of thermodynamic binding affinitymust also consider stoichiometry. The interaction of a singleFab domain with its cognate antigen generally represents a1:1 or ‘‘monovalent’’ protein–protein interaction in whichEq. (1) applies. However, for divalent proteins, such as full-length antibodies or Fab02 fragments with two antigen-combining sites, the observed affinity depends heavily onthe conditions under which the binding constant is measured.In situations where the affinity is determined by measuringbinding of the antibody to a surface coated with antigen, suchas SPR, enzyme-linked immunosorbent assay (ELISA), fluor-escence-activated cell sorting (FACS), or whole-cell bindingassays, the observed affinity can be significantly enhancedby avidity effects, as simultaneous dissociation of both anti-gen-binding regions from surface-immobilized antigen occursless frequently than single dissociation events (36,37). Thiseffect can be quite advantageous for antibodies that bindmembrane bound antigens. In fact, significant research hasbeen focused on exploiting the avidity effect by generatingpolyvalent antibodies with four, six, or eight antigen-bindingregions (38). However, as avidity effects are necessarilyhighly dependent upon assay conditions (36,37), we will dis-cuss affinity determinations here in the context of the 1:1binding affinity of an antibody’s antigen-combining site witha single site on its antigen. The measurement of this monova-lent interaction may be made in the context of the bivalentIgG if care is taken to prevent bivalent interactions; seefor example Ref. 35.

In most clinical applications of antibodies, the formationof a complex between the antibody and its cognate antigen isessential if not synonymous to its activity. The field of in vitroaffinity maturation has developed to improve upon thepotency and efficacy of many first-generation human, huma-nized, and nonhuman antibodies that may be limited by anti-gen-binding affinity. Enhanced binding affinity as measuredin vitro has proved beneficial to potency in vivo (39).

During in vitro affinity maturation, an antibody issubjected to mutagenesis and selection with the hope thatan ‘‘improved’’ antibody lies near the parent antibody in


sequence space. Because it is impossible to sample all poten-tial sequences, extensive effort has been focused on develop-ing methods for constructing more efficient libraries andbetter methods for selecting high-affinity binders from them.Here, we review the literature on in vitro affinity maturationusing phage display (see Table 2), as well as other displaytechniques including cell-surface display and ribosome dis-play, and compare the strategies and methods employed.

In designing any in vitro affinity maturation experiment,the key considerations are how large the library can andshould be, which amino acid positions should be randomized(specific positions, one particular CDR, all CDRs, just theheavy or light chain), by what technique genetic diversity willbe generated (oligonucleotide-directed mutagenesis, error-prone PCR, DNA shuffling), and how much amino acid diver-sity will be allowed at each randomized position (all 20 aminoacids, only hydrophilic residues, only amino acids resultingfrom single point mutations). Additionally, one must considerthe structural format for antibody display (Fab, scFv) andhow the library will be selected (or ‘‘panned’’) for higher-affinity binders (solution binding and capture, immobilizedantigen).

As seen in Fig. 3, successes in the in vitro affinitymaturation of antibodies reported span a broad range of abso-lute affinities (from 1mM to 48 fM) and relative improvementsin affinity (from two- to 6250-fold improvement). The startingaffinities of these antibodies for their antigens have rangedfrom subnanomolar to nearly 50mM, with examples of hun-dreds-fold improvement in each affinity range (Table 2). Withmore than 20 reports of antibody affinity maturation studies,we can begin to compare the approaches and experimentalmethods for their effectiveness in identifying tighter bindingantibodies.

III.A. Library Size

The central problem in affinity maturation using combinator-ial diversity is to efficiently select from among the astronom-ical number of possible amino acid sequences at least one


Table 2 Reports of In Vitro Affinity Maturation Surveyed in this Review

Reference AntigenStartingKd (nM)

EndingKd (nM)

Improved(fold)

Displayformat

Mutagenmethod

Mutanttargeting

Maximumtheoreticaldiversity

Maximumsampleddiversity

49 Fluoresceina 0.3 4.8 � 10�5 6,250 scFv Mixed Random NA 2� 107

60 HER2 16 0.013 1,231 scFv Oligo Function 5.1�1011 1� 107

48 Gp120 6.3 0.015 420 Fab Oligo CDRs 6.4�107 6.7�107

53 IL-1b 0.5 0.050 10 Fab Oligo CDRs 1.2�102 1.6�105

20 VEGF 13 0.11 118 Fab Oligo Structure 3.2�106 3.2�106

61 gp120 2 0.2 10 scFv Chain shuffling CDRs NA 3� 106

62 Mesothelin 11 0.2 55 scFv Oligo CDRsc 8�103 8� 103

54 Digoxinb 0.9 0.3 3 scFv Oligo Structure 1.6�105 1.6�105

41 avb3 28 0.3 92 Fab Oligo CDRs 2.6�103 2.6�103

42 EpCam 6 0.4 15 ScFv Chain shuffling Random NA 6� 107

63 phOx 320 1.1 291 ScFv Chain shuffling CDRs NA 2.2�106

64 Fibronectin 110 1.1 100 ScFv Oligo CDRs NA NA65 EGFR vIII 9 2 5 ScFv Oligo CDRs 8�103 8� 103

66 NIP-caproic 42 9 5 ScFv PCRd Random NA 4� 104

67 Glycophorin 48,000 100 480 ScFv Mutator strain Random NA 1� 1013

68 preS1 of HBV 8,000 230 35 ScFv Chain shuffling Random NA 3.2�108

47 Ars 3,600 600 6 Fab Oligo Function 400 40069 LewisY antigen 9,900 700 14 Fab Oligo CDRs NA 1� 108

70 Progesterone 30,000 1,000 30 Fab scFv PCRd Random NA 5� 106

aYeast surface display.bBacteria surface display.cUsed DNA hot-spots to target mutagenesis.dUsed error-prone PCR for mutagenesis.

Antib

odyHuman

izationan

dAffinity

Matu

rationUsin

gPhage

Disp

lay507

sequence that is optimal for any given application. It isinferred that the likelihood of discovering a better binder willincrease with number of sequences sampled (40). Althoughthere is a clear advantage to sampling more sequence space,a larger library does not guarantee that the highest affinitybinders will be isolated. For instance, in one case (41), a 92-fold-improved, subnanomolar antibody was obtained from alibrary of just 2592 sequences, while in another (42), a libraryof 6 � 107 sequences improved binding only 15-fold.

To evaluate whether larger libraries more generallygenerate tighter binders, we have plotted both these values(Fig. 4a,b) as a function of the maximum number of sequencessampled for each reported study. Figure 4b shows that theremay be some general trends towards increasing improve-ments in affinity with increasing library size. However, thereis no obvious correlation between absolute affinity and librarysize (Fig. 4a).

Figure 3 The results of antibody affinity maturation vary widely,in both overall affinity (48 fM for fluorescein–biotin, 1 mM for pro-gesterone) and magnitude of improvement (6250-fold for fluores-cein–biotin, threefold for digoxin). (Compiled from Refs. 49, 54, 70).


Also revealed by Fig. 4 is the fact that there is no clear dis-tinction between the successes of libraries that oversample orundersample the theoretical maximum diversity of thelibrary. Oversampling has the advantage of providing valu-able information in the event that a better binder is notretrieved from the library—one is confident of the exact regionof sequence space around the parental sequence that has beensearched, and that higher-affinity antibodies are unlikely toarise from searching that space again. Unfortunately, thisrequires the spectrum of diversity to be predetermined, andthus is limited by the biases inherent to the design of theexperiment. Undersampling has the advantage of accessingan extremely broad and unbiased spectrum of diversity, buthas no guarantee that repeating the selection will yield thesame (or even similar) results. Thus, little is learned by obtain-ing a negative result from poorly sampled libraries.

Figure 4 Relationship between library size and (a) absolute affi-nity or (b) affinity improvement relative to the parental antibodyfragment. For oversampled libraries (empty circles), the sampleddiversity is the maximum theoretical diversity for mutagenic oligo-nucleotides used. For undersampled libraries (filled circles), thesampled diversity is the constructed library size. The librarywith 1013 members was constructed by passage through a mutatorstrain. (From Ref. 67.)


III.B. Targeting Positions for RandomMutagenesis

Another critical facet of the library design process is deter-mining which amino acid positions should be randomized(Fig. 5). Intuitively, one would expect that targeting positionsthat make direct contacts with the antigen would be morelikely to result in significant improvements in affinity thanwould targeting positions that are distant from the antigen-binding site. Early affinity maturation studies indicated thatboth residues intimately involved in intermolecular contactsas well as those at the periphery of the interface could yieldaffinity improvements (43). If a structure of the antibody–antigen complex is available, identifying these contactresidues is straightforward, as in the case for an anti-VEGFantibody, which was affinity-matured by a factor of >20-fold(20). In this case, a total of four CDR changes, identified bytargeting contact and noncontact residues, resulted in theimproved variant Y0317 (see Fig. 2). This variant, also knownas Lucentis2 in Fab form, is being investigated for the treat-ment of macular degeneration (44).

Without a structure, the library design can be guided byselecting surface-exposed CDR residues in general, or by iden-tifying residues important for binding (in another applicationof antibody–phage display) via site-directed (45) or shotgun(46) alanine-scanning mutagenesis of the CDRs. Theseapproaches have produced targeted libraries that resulted ina number of successful affinity optimizations (47). It is alsopossible to improve affinity by ‘‘walking though’’ the CDRs(48), limiting mutations to one CDR at a time and combiningthe best mutations from each.

Alternatively, significant increases in affinity can beachieved by constructing a library, either in the heavy chain,the light chain, or the entire antibody gene, without any spe-cific targeting. In fact, to our knowledge, the highest affinityantibody reported (and the one with the greatest improve-ment in affinity) was isolated by an untargeted mutagenesisscheme (49), with mutations important for increasing affinitylocated as far as 25 A away in a ‘‘fourth shell’’ surrounding the


direct contact residues. All of these methods show varyingdegrees of success, as shown in Fig. 6, with no single methodclearly surpassing others. With respect to therapeutic antibo-dies, however, limiting the sites of mutagenesis to normally

Figure 5 Location of mutations identified by in vitro affinitymaturation in the variable domain of various antibodies. Top, ‘‘sideview.’’ Bottom, ‘‘bird’s eye view’’ of binding interface. Beneficialmutations identified in the reports surveyed here are mapped ontothe structure of a humanized Fv (1FCV.pdb). The light-chain (VL;gray) and heavy-chain (VH;black) components of the variable domainare shown in ribbon form. Mutations obtained through oligonucleo-tide-directed mutagenesis, primarily within the CDRs, are shown asdark gray spheres. Mutations obtained through undirected, randommutagenesis are shown as light gray spheres. Image constructedwith PyMol. (From Ref. 71.)


hypervariable residues (50) has the presumed advantage ofreducing the risk of immunogenicity in selected variantswithout the necessity of testing the effects of mutations atconserved sites.

III.C. Method of Mutagenesis

As mentioned previously, one must also consider whether theconstructed library will over- or undersample the theoreticalmaximum diversity of the library. For a statistical analysisof the degree to which the actual library size must exceedthe theoretical maximum library size, see Ref. 51. Over-sampled libraries can be obtained by oligonucleotide-directedmutagenesis (either template- or cassette based) at arestricted number of sites. Oligonucleotide-directed mutagen-esis also allows the experimenter to control exactly where

Figure 6 Relationship between (a) absolute affinity or (b) affinityimprovement relative to the parental antibody fragment and selec-tion of target residues for mutagenesis. Two studies used structuralinformation (Struct.) to guide mutagenesis (Refs. 20,54) and twostudies used functional analysis (Funct.) (Refs. 47,60). The remain-ing studies either targeted CDR residues in general (CDRs) or intro-duced random mutations (Random) throughout the molecule viaPCR or mutator strain.


mutations are introduced, thus alleviating concern aboutintroducing potentially immunogenic mutations to frameworkresidues. Any nonsystematic mutagenesis method that intro-duces mutations at completely random positions (e.g., via pas-sage through a mutator strain, error-prone PCR, DNA‘‘shuffling’’) requires no information on the importance of eachresidue to binding, and permits the maturation of antibodieswithout the imposed biases of the experimenter. However,such methods also increase the theoretical maximum diver-sity to the point that it cannot be effectively sampled.Although some of these methods can limit mutagenesis to spe-cific regions of amino acid sequence (e.g., by using error-pronePCR to randomly mutate just one or two CDRs), the technicalexpertise required can offset the simplicity and ease thatmakes random mutagenesis so attractive.

The technique used to introduce mutations is highlycoupled to the position-targeting scheme. However, whetheroligonucleotide-based, PCR, or other methods are used, themutagenesis method alone is not correlated with the successof the antibody selection (Fig. 7).

III.D. Diversity and Codon Degeneracy

In vivo affinity maturation relies on somatic mutation andrecombination of the CDRs, yet all natural antibodies origi-nate from a relatively small set of germline variable-domaingenes which themselves have arisen from evolutionary geneduplication. There are two schools of thought on how closelythe in vitro affinity maturation process should parallel thein vivo one. One perspective is that if antibodies with highaffinity are derived in vivo by somatic mutation, which intro-duces primarily single-nucleotide point mutations (and nottriple-nucleotide codon replacements), then high-affinityantibodies should be selected in vitro from libraries them-selves constructed by methods that introduce primarily sin-gle-nucleotide point mutations. Another perspective is thatthere is no intrinsic thermodynamic reason for the geneticdiversity of an in vitro library to be biased by the evolutionaryhistory of the genetic code, and thus all 20 amino acids should


comprise the library. Although both perspectives are based oncompelling hypotheses, a survey of reported improvements inaffinity shows that significant improvements in affinity can beobtained through either approach (Fig. 8).

III.E. Antigen-Binding Selection Methods

Once the library is designed, constructed, and transformedinto E. coli (as described in earlier chapters), it must be sub-jected to antigen-binding selections for antibodies with higheraffinity than the parent. Depending on how the library was

Figure 7 Relationship between mutagenesis method and (a) abso-lute affinity or (b) affinity improvement relative to the parentalantibody fragment. Methods are described in the text. ‘‘Multiplemethods’’ refers to a combination of mutator strains, chain shufflingand PCR. For oversampled libraries (empty circles), the sampleddiversity is the maximum theoretical diversity for mutagenic oligo-nucleotides used. For undersampled libraries (filled circles), thesampled diversity is the size of the library actually constructed.(From Ref. 49.)


designed, it is possible that a significant number of the var-iants may be very similar to wild type, making it essentialthat the selections be performed under conditions that canfinely discriminate between antibodies with similar affinities.The most common technique, panning for binders with sur-face immobilized antigen (2), has provided very impressiveresults (Fig. 9). In this technique, a 96-well plate (usuallymodified polystyrene or polycarbonate) is coated with antigenat 1–10 mg=mL and blocked with BSA, casein, or anotherblocking reagent to prevent adsorption of phage. The phagelibrary (�1010–1012 phage=mL) is allowed to bind for a periodof minutes to hours and then nonbinding and weaker binding

Figure 8 Relationship between (a) absolute affinity or (b) affinityimprovement relative to the parental antibody fragment and anycodon bias that may result from the mutagenesis method. Innonoligonucleotide-directed mutagenesis (error-prone PCR or shuf-fling), it is assumed that most amino acids changes arise fromsingle-nucleotide point mutations (Single Mut.), and thus will bebiased towards amino acids that are more closely related in thegenetic code to the wild-type amino acid at any given position.Oligonucleotide-directed mutagenesis incorporates amino acidswith only the bias of the degeneracy of the genetic code (Full Mut.).


phage are washed off. To discriminate among many phageswith high affinities and isolate phages with the slowest off-rates, washing can take place over a period of hours (or evendays) with soluble antigen in the wash buffer to preventrebinding. Although this technique is simple to implement,it takes experience to know how to adjust the washing para-meters to get the best binders.

Another selection method is solution sorting (52). In thismethod, the phage-displayed library is incubated with solubleantigen that has been tagged with biotin. Bound phages arecaptured on streptavidin-coated wells or beads, washed,eluted, and propagated. By controlling the concentration ofantigen, one can control the stringency of the selection. Forexample, selecting with 1nM antigen should favor recoveryof phages with a Kd of 1 nM or lower.

Figure 9 Relationship between (a) absolute affinity or (b) affinityimprovement relative to the parental antibody fragment and theselection method used. Most selections were performed on a solidsupport (‘‘Solid,’’ indicating intact cells or antigen immobilized ona plastic surface), while some used a solution binding and capturetechnique (Sol’n) or relied on screening individual variants(Screen).


It is also possible to identify affinity-improved variantsby simply screening each clone. This method precludes thetechnical intricacies inherent to display and ensures thatrare, high-affinity clones will not be lost or out grown duringthe propagation step used in phage display. However, itrequires either a very small library or high throughputscreening machinery. The most ‘‘successful’’ implementationof this technique is the affinity maturation of an anti avb3antibody 92-fold (41). By screening all possible single aminoacid mutations at each CDR position (2336 mutants), Wu etal. were able to identify individual mutations that increasedaffinity two- to 13-fold and then combinatorially combinethose (256 mutants) and identify multiple clones with evenhigher affinity. Similarly, but not as dramatically, Cassonand Manser (47) improved the affinity of an antiarsenateantibody by sixfold by screening all 400 possible mutationsat two amino acid positions.

III.F. Merging Results from Separate Libraries

In a number of cases using targeted mutagenesis schemes, invitro affinity maturation experiments have used multiplelibraries to sample mutations in each CDR. A key questionthat must then be addressed is how to combine the mutationsderived from each library. There are three basic approachesone can take (Fig. 10). One option is to construct a numberof separate libraries, identify critical mutations in each, andassume that their effects on affinity will be roughly additive.Sometimes this approach is successful (20,53), but it is naıveto think it always will be, as the intricacies of protein–proteininteractions are not yet completely understood. Anotheroption is to proceed serially, sometimes called ‘‘CDR walking’’(48), making a library of one CDR, selecting the best binder,and using that improved variant as the framework for thenext library with different amino acid positions randomized.This approach has the obvious advantage that binders fromeach successive library are guaranteed to work within thecontext of the previously identified beneficial mutations. How-ever, synergistic mutations bridgingmultiple CDRs will not beidentified by this technique. A third option, which addresses


the inadequacies of the assumptions of additivity in the firstdescribed approach, is more parallel in nature, with separatelibraries being constructed and sorted for each CDR (or othertargeted mutagenic region), and then recombined combinato-rially to identify synergistically interacting residues (41).

III.G. Antibody Format

Another factor to consider in an affinity maturation experi-ment is the antibody format. Both antigen-binding fragments(Fabs), usually with the heavy-chain VH–CH1 domains fusedto the phage protein and VL–CL coexpressed as a soluble sec-ond chain, and single-chain variable fragments (scFv), withVH and VL domains connected via a peptide linker, have been

Figure 10 Relationship between (a) absolute affinity or (b) affi-nity improvement relative to the parental antibody fragment andthe means by which mutations from separate selections or panningswere recombined to yield yet higher-affinity antibodies. The major-ity of studies performed only one panning experiment (None).Others used the winners from one panning as the framework fora second panning (Serial). Two cases of simple additivity werereported (Add.), as were two cases of making an additional librarycontaining the best mutations from previous libraries followed byadditional selections (Parallel).


used for phage display. The single-chain format may haveexpression advantages in general; however, as discussedabove, the formation of diabodies can produce misleadingresults in binding and activity assays. Both formats continueto be widely used, and neither has demonstrated a clearadvantage over the other with respect to in vitro affinitymaturation (Fig. 11).

IV. EMERGING APPROACHES

Recently, two additional display technologies have been usedfor affinity maturation: FACS and ribosome display.Fluorescence-activated cell sorting separates populations ofcells based on their fluorescent properties. This technologyhas been used to affinity mature an antifluorescein–biotinantibody by display on the surface of yeast (49), resulting inthe greatest improvement in affinity (6250-fold) and the

Figure 11 Relationship between display format (Fab or scFv) and(a) absolute affinity or (b) affinity improvement relative to theparental antibody fragment.


tightest overall antibody (48 fM). Fluorescence-activatedcell sorting has also been used to separate bacterial cells, dis-playing scFv via fusion with the OmpA bacterial surfaceprotein, though resulting in only minor affinity improvements(54). One obvious limitation of this technology is that itdepends on the use of a soluble fluorescent target.

Ribosomal display is a completely ex vivo approach inwhich the protein of interest is tethered to the actual ribo-some (and mRNA) that encodes it (55,56). Although this tech-nology has not been used to affinity mature an antibody perse, but rather isolate antibodies with picomolar affinity(57,58), it is likely that this technology will be adapted soonfor affinity maturation.

In addition to these new methods for sorting=selectingimproved binders from large libraries of variants, progresshas been made in the use of computational methods forimproving affinity. In the case of anti-VEGF antibodies,variants with improved affinity were designed by specificallytargeting residues within and outside the set of antigen-contacting residues (59). In this case, although extensivephage selection strategies had led to improved affinitythrough off-rate selections, no significant improvements inon-rate had been found. The computational approach led toa new set of mutations that provide enhanced affinity primar-ily through enhanced on-rate in the ‘‘34-TKKT’’ variant (seeFig. 2).

V. CONCLUSIONS

With many examples of humanization and affinity matura-tion of antibody fragments in the literature, one can beginto compare the methods employed to address the variablesof display-based affinity maturation. Since a variety of muta-genesis, residue-targeting, and selection methods have beenused, comparisons between specific methodologies can onlybe made with a limited number of examples of each. However,the picture that emerges at this point is that no ‘‘magic bullet’’(to borrow a long-used reference to therapeutic antibodies)


has been found for systematically improving the affinity ofantibodies, but rather each method has its own benefits andlimitations. The success of a variety of approaches likelyarises from the fact that multiple high-affinity solutions oftenexist for optimized protein–protein interfaces, and the factthat single amino acid changes (or the combination of a fewchanges) can often have dramatic effects on binding affinity.

Ultimately, the combination of structure-based design,structure-based diversity design, and stringent selection stra-tegies are likely to yield the most useful antibodies for reagentand therapeutic use.

REFERENCES

1. Kohler G,Milstein C. Continuous cultures of fused cells secretingantibody of predefined specificity. Nature 1975; 256:495–497.

2. Smith GP. Filamentous fusion phage: novel expression vectorsthat display cloned antigens on the virion surface. Science1985; 228:1315–1317.

3. Chothia C, Lesk AM. Canonical structures for the hypervari-able regions of immunoglobulins. JMol Biol 1987; 196:901–917.

4. Wilson IA, Stanfield RL. Antibody–antigen interactions: newstructures and new conformational changes. Curr Opin StructBiol 1994; 4:857–867.

5. Kang AS, Barbas CF III, Janda KD, Benkovic SJ, Lerner RA.Linkage of recognition and replication functions by assemblingcombinatorial antibody Fab libraries along phage surfaces.Proc Natl Acad Sci USA 1991; 88:4363–4366.


7. Barbas CF III, Kang AS, Lerner RA, Benkovic SJ. Assembly ofcombinatorial antibody libraries on phage surfaces: the geneIII site. Proc Natl Acad Sci USA 1991; 88:7978–7982.

8. Hoogenboom HR, Griffiths AD, Johnson KS, Chiswell DJ,Hudson P, Winter G. Multi-subunit proteins on the surface of


filamentous phage: methodologies for displaying antibody (Fab)heavy and light chains. Nucleic Acids Res 1991; 19:4133–4137.

9. Marks JD, Hoogenboom HR, Bonnert TP, McCafferty J,Griffiths AD, Winter G. By-passing immunization. Humanantibodies from V-gene libraries displayed on phage. J MolBiol 1991; 222:581–597.

10. Perisic O, Webb PA, Holliger P, Winter G, Williams RL.Crystal structure of a diabody, a bivalent antibody fragment.Structure 1994; 2:1217–1226.

11. Jaffers GJ, Fuller TC, Cosimi AB, Russell PS, Winn HJ, ColvinRB. Monoclonal antibody therapy. Anti-idiotypic and non-anti-idiotypic antibodies to OKT3 arising despite intenseimmunosuppression. Transplantation 1986; 41:572–578.

12. Riechmann L, Clark M, Waldmann H, Winter G. Reshapinghuman antibodies for therapy. Nature 1988; 332:323–327.

13. Hale G, Dyer MJ, Clark MR, Phillips JM, Marcus R,Riechmann L, Winter G, Waldmann H. Remission inductionin non-Hodgkin lymphoma with reshaped human monoclonalantibody CAMPATH-1H. Lancet 1988; 2:1394–1399.

14. Sugo T, Mizuguchi J, Kamikubo Y, Matsuda M. Anti-humanfactor IX monoclonal antibodies specific for calcium ion-induced conformations. Thromb Res 1990; 58:603–614.

15. Suggett S, Kirchhofer D, Hass P, Lipari T, Moran P, Nagel M,Judice K, Schroeder K, Tom J, Lowman H, Adams C, Eaton D,Devaux B. Use of phage display for the generation of humanantibodies that neutralize factor IXa function. Blood CoagulFibrinolysis 2000; 11:27–42.

16. Carter P, Presta L, Gorman CM, Ridgway JB, Henner D,Wong WL, Rowland AM, Kotts C, Carver ME, Shepard HM.Humanization of an anti-p185HER2 antibody for humancancer therapy. Proc Natl Acad Sci USA 1992; 89:4285–4289.

17. Presta LG, Lahr SJ, Shields RL, Porter JP, Gorman CM,Fendly BM, Jardieu PM. Humanization of an antibody direc-ted against IgE. J Immunol 1993; 151:2623–2632.

18. Werther WA, Gonzalez TN, O’Connor SJ, McCabe S, Chan B,Hotaling T, Champe M, Fox JA, Jardieu PM, Berman PW,Presta LG. Humanization of an anti-lymphocyte function-


associated antigen (LFA)-1 monoclonal antibody and reengi-neering of the humanized antibody for binding to rhesusLFA-1. J Immunol 1996; 157:4986–4995.

19. Presta LG, Chen H, O’Connor SJ, Chisholm V, Meng YG,Krummen L, Winkler M, Ferrara N. Humanization of ananti-vascular endothelial growth factor monoclonal antibodyfor the therapy of solid tumors and other disorders. CancerRes 1997; 57:4593–4599.

20. Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P,de Vos AM, LowmanHB. Selection and analysis of an optimizedanti-VEGF antibody: crystal structure of an affinity-maturedFab in complex with antigen. J Mol Biol 1999; 293:865–881.

21. Foote J, Winter G. Antibody framework residues affecting theconformation of the hypervariable loops. J Mol Biol 1992;224:487–499.

22. Jones PT, Dear PH, Foote J, Neuberger MS, Winter G. Repla-cing the complementarity-determining regions in a humanantibody with those from a mouse. Nature 1986; 321:522–525.

23. Verhoeyen M, Milstein C, Winter G. Reshaping human antibo-dies: grafting an antilysozyme activity. Science 1988; 239:1534–1536.

24. Chothia C, Lesk AM, Tramontano A, Levitt M, Smith-Gill SJ,Air G, Sheriff S, Padlan EA, Davies D, Tulip WR, Colman PM,Spinelli S, Alzari PM,PoljakRJ. Conformations of immunoglob-ulin hypervariable regions. Nature 1989; 342:877–883.

25. Gussow D, Seemann G. Humanization of monoclonal antibo-dies. Methods Enzymol 1991; 203:99–121.

26. Carter P, Abrahmsen L, Wells JA. Probing the mechanism andimproving the rate of substrate-assisted catalysis in subtilisinBPN. Biochemistry 1991; 30:6142–6148.

27. Baca M, Presta LG, O’Connor SJ, Wells JA. Antibody humani-zation using monovalent phage display. J Biol Chem 1997;272:10678–10684.

28. Queen C, Schneider WP, Selick HE, Payne PW, Landolfi NF,Duncan JF, Avdalovic NM, Levitt M, Junghans RP,Waldmann TA. A humanized antibody that binds to the


interleukin 2 receptor. Proc Natl Acad Sci USA 1989; 86:10029–10033.

29. Rosok MJ, Yelton DE, Harris LJ, Bajorath J, Hellstrom KE,Hellstrom I, Cruz GA, Kristensson K, Lin H, Huse WD,Glaser SM. A combinatorial library strategy for the rapidhumanization of anticarcinoma BR96 Fab. J Biol Chem 1996;271:22611–22618.

30. Roguska MA, Pedersen JT, Keddy CA, Henry AH, Searle SJ,Lambert JM, Goldmacher VS, Blattler WA, Rees AR,Guild BC. Humanization of murine monoclonal antibodiesthrough variable domain resurfacing. Proc Natl Acad SciUSA 1994; 91:969–973.

31. Jespers LS, Roberts A, Mahler SM,Winter G, HoogenboomHR.Guiding the selection of human antibodies from phage displayrepertoires to a single epitope of an antigen. Biotechnology1994; 12:899–903.

32. Ferrara N. Role of vascular endothelial growth factor in phy-siologic and pathologic angiogenesis: therapeutic implications.Semin Oncol 2002; 29:10–14.

33. Klotz IM. Ligand–receptor interactions: facts and fantasies. QRev Biophys 1985; 18:227–259.

34. Neri D, Montigiani S, Kirkham PM. Biophysical methods forthe determination of antibody–antigen affinities. Trends Bio-technol 1996; 14:465–470.

35. Karlsson R, Michaelsson A, Mattsson L. Kinetic analysis ofmonoclonal antibody–antigen interactions with a new biosen-sor based analytical system. J Immunol Methods 1991;145:229–240.

36. Crothers DM, Metzger H. The influence of polyvalency on thebinding properties of antibodies. Immunochemistry 1972;9:341–357.

37. Muller KM, Arndt KM, Pluckthun A. Model and simulation ofmultivalent binding to fixed ligands. Anal Biochem 1998;261:149–158.

38. Miller K, Meng G, Liu J, Hurst A, Hsei V, Wong WL, Ekert R,Lawrence D, Sherwood S, DeForge L, Gaudreault J, Keller G,Sliwkowski M, Ashkenazi A, Presta L. Design, construction,


and in vitro analyses of multivalent antibodies. J Immunol2003; 170:4854–4861.

39. Schlom J, Eggensperger D, Colcher D,Molinolo A, Houchens D,Miller LS, Hinkle G, Siler K. Therapeutic advantage ofhigh-affinity anticarcinoma radioimmunoconjugates. CancerRes 1992; 52:1067–1072.

40. Perelson AS. Immune network theory. Immunol Rev 1989;110:5–36.

41. Wu H, Beuerlein G, Nie Y, Smith H, Lee BA, Hensler M,Huse WD, Watkins JD. Stepwise in vitro affinity maturationof Vitaxin, an avb3-specific humanized mAb. Proc Natl AcadSci USA 1998; 95:6037–6042.

42. Huls G, Gestel D, van der Linden J, Moret E, Logtenberg T.Tumor cell killing by in vitro affinity-matured recombinanthuman monoclonal antibodies. Cancer Immunol Immunother2001; 50:163–171.

43. Lowman HB, Wells JA. Affinity maturation of human growthhormone by monovalent phage display. J Mol Biol 1993;234:564–578.

44. N Ferrara. VEGF and the quest for tumour angiogenesis fac-tors. Nat Rev Cancer 2002; 2:795–803.

45. Muller YA, Chen Y, Christinger HW, Li B, Cunningham BC,Lowman HB, de Vos AM. VEGF and the Fab fragment of ahumanized neutralizing antibody: crystal structure of the com-plex at 2.4 A resolution and mutational analysis of the inter-face. Structure 1998; 6:1153–1167.


47. Casson LP, Manser T. Random mutagenesis of two comple-mentarity determining region amino acids yields an unexpect-edly high frequency of antibodies with increased affinity forboth cognate antigen and autoantigen. J Exp Med 1995;182:743–750.

48. Yang WP, Green K, Pinz-Sweeney S, Briones AT, Burton DR,Barbas CF III. CDR walking mutagenesis for the affinity


maturation of a potent human anti-HIV-1 antibody into thepicomolar range. J Mol Biol 1995; 254:392–403.

49. Boder ET, Midelfort KS, Wittrup KD. Directed evolution of anti-body fragments with monovalent femtomolar antigen-bindingaffinity. Proc Natl Acad Sci USA 2000; 97:10701–10705.

50. Kabat EA, Wu TT, Perry HM, Gottesman KS, Foeller C.Sequences of Proteins of Immunological Interest. Bethesda,MD: National Institutes of Health, 1991.

51. Lowman HB, Wells JA. Monovalent phage display: a methodfor selecting variant proteins from random libraries. Methods:Companion Methods Enzymol 1991; 3:205–216.


53. Jackson JR, Sathe G, Rosenberg M, Sweet R. In vitro antibodymaturation. Improvement of a high affinity, neutralizing anti-body against IL-1 beta. J Immunol 1995; 154:3310–3319.

54. Daugherty PS, Chen G, Olsen MJ, Iverson BL, Georgiou G.Antibody affinity maturation using bacterial surface display.Prot Eng 1998; 11:825–832.

55. Mattheakis LC, Bhatt RR, Dower WJ. An in vitro polysomedisplay system for identifying ligands from very largepeptide libraries. Proc Natl Acad Sci USA 1994; 91:9022–9026.

56. Hanes J, Pluckthun A. In vitro selection and evolution of func-tional proteins by using ribosome display. Proc Natl Acad SciUSA 1997; 94:4937–4942.

57. Hanes J, Jermutus L, Weber-Bornhauser S, Bosshard HR,Pluckthun A. Ribosome display efficiently selects and evolveshigh-affinity antibodies in vitro from immune libraries. ProcNatl Acad Sci USA 1998; 95:14130–14135.

58. Hanes J, Schaffitzel C, Knappik A, Pluckthun A. Picomolar affi-nity antibodies from a fully synthetic naive library selected andevolved by ribosome display. Nat Biotechnol 2000; 18:1287–1292.

59. Marvin JS, Lowman HB. Redesigning an antibody fragmentfor faster association with its antigen. Biochemistry 2003;42:7077–7083.


60. Schier R, Marks JD. Efficient in vitro affinity maturation ofphage antibodies using BIAcore guided selections. Hum Anti-bodies Hybridomas 1996; 7:97–105.

61. Thompson J, Pope T, Tung JS, Chan C, Hollis G, Mark G,Johnson KS. Affinity maturation of a high-affinity humanmonoclonal antibody against the third hypervariable loop ofhuman immunodeficiency virus: use of phage display toimprove affinity and broaden strain reactivity. J Mol Biol1996; 256:77–88.

62. Chowdhury PS, Pastan I. Improving antibody affinity bymimicking somatic hypermutation in vitro. Nat Biotechnol1999; 17:568–572.

63. Marks JD, Griffiths AD, Malmqvist M, Clackson TP, Bye JM,Winter G. By-passing immunization: building high affinityhuman antibodies by chain shuffling. Bio-Technology 1992;10:779–783.

64. Neri D, Carnemolla B, Nissim A, Leprini A, Querze G, Balza E,Pini A, Tarli L, Halin C, Neri P, Zardi L, Winter G. Targetingby affinity-matured recombinant antibody fragments of anangiogenesis associated fibronectin isoform. Nat Biotechnol1997; 15:1271–1275.

65. Beers R, Chowdhury P, Bigner D, Pastan I. Immunotoxinswith increased activity against epidermal growth factor recep-tor vIII-expressing cells produced by antibody phage display.Clin Cancer Res 2000; 6:2835–2843.

66. Hawkins RE, Russell SJ, Winter G. Selection of phage antibo-dies by binding affinity. Mimicking affinity maturation. J MolBiol 1992; 226:889–896.

67. Irving RA, Kortt AA, Hudson PJ. Affinity maturation ofrecombinant antibodies using E. coli mutator cells. Immuno-technology (Amsterdam) 1996; 2:127–143.

68. Park SG, Lee JS, Je EY, Kim IJ, Chung JH, Choi IH. Affinitymaturation of natural antibody using a chain shuffling tech-nique and the expression of recombinant antibodies in Escher-ichia coli. Biochem Biophys Res Commun 2000; 275:553–557.

69. Yelton DE, Rosok MJ, Cruz G, Cosand WL, Bajorath J,Hellstrom I, Hellstrom KE, Huse WD, Glaser SM. Affinity


maturation of the BR96 anti-carcinoma antibody by codon-based mutagenesis. J Immunol 1995; 155:1994–2004.

70. Gram H, Marconi LA, Barbas CF III, Collet TA, Lerner RA,Kang AS. In vitro selection and affinity maturation of anti-bodies from a naive combinatorial immunoglobulin library.Proc Natl Acad Sci USA 1992; 89:3576–3580.

71. Delano WL. The PyMOL Molecular Graphics System. DeLanoScientific, 2002.


15

Antibody Libraries from ImmunizedRepertoires

JODY D. BERRY

Monoclonal Antibody Section,Reagent Development Unit, NationalCentre for Foreign Animal Disease,

Winnipeg, Manitoba, Canada

MIKHAIL POPKOV

Department of Molecular Biology,The Scripps Research Institute,

La Jolla, California, U.S.A.

I. INTRODUCTION

Polyclonal antibodies have been used from diverse speciessuch as rabbit, goat, sheep, donkey, chicken, pigs, cats, dogs,minks, and cattle for almost a century as specific and high-affinity probes for a variety of immunological assays inresearch and clinical laboratories. While polyclonal-pooledimmunoglobulin (Ig) is still the gold standard for the treat-ment of some human diseases, the high purity and specificityof monoclonal antibody will eventually oust these serum pre-parations from this position; monoclonal antibodies (MAbs)promise to be safer and more specific in therapy. While there

529

are a few examples where MAbs have been developed for all ofthe above species (1,2), MAb development in these species hasbeen very slow mostly because of a lack of reliable myelomafusion partners resulting in unstable clones and complex backfusions to produce heterohybridomas (1). Repertoire cloningby phage display has offered a new method for obtainingMAbs from immunized repertoires from all of these species,and for the immune systems of all jawed vertebrates in thefuture.

Antibody libraries represent a major means by whichMAbs are produced today. Monoclonal antibodies can be madeeasily and reproducibly in large quantities without lot varia-tion seen in polyclonal antibody preparations, therefore MAbsallow many experiments to be performed which were not pre-viously possible or practical. The use of MAbs avoids undesir-able cross-reactivity and high-affinity MAbs can be achievedeasily. The purity of MAbs produced in vitro is extremelyhigh. The uniform nature of pure MAbs has led to widespreaduse in biotechnology and biopharmaceutical research. Mono-clonal antibodies produced in vitro also have less risk of carry-ing unknown passenger viruses and thus are safer comparedto pooled human gamma globulin preparations of MAbs fromhumans for use in therapy. Other applications include:research tools for cell, antigen, or pathogen identification;ligands for column chromatography and molecule purifica-tion; as diagnostic reagents; therapeutic antibody prepara-tions and vaccine development. This is in addition to themany procedures in which MAbs are used in the basic sciencelab. Clearly MAbs have indisputable value in modern biologi-cal laboratories. Antibody libraries, to date, have had theirgreatest application in the development of therapeutic MAbsin humans. In many ways, the lure of therapeutic MAbs hasdriven the field of immune libraries and it is only morerecently that experiments on novel animal species have beeninitiated.

The antibody library approach provides a recombinantmeans to derive MAbs from various sources in vitro. Thelibraries themselves allow scientists to easily obtain thegenetic information that encodes a given MAb. This has been

530 Berry and Popkov

useful, in itself, for elucidating fundamental immunologicalprinciples and has contributed to the knowledge of antibodygenetics. The three common formats of antibody library avail-able include immune, naıve, and synthetic libraries (3,4).Immune libraries are made from the antigen sensitized B cellIgG mRNA. While the antibody fragments selected fromimmune libraries are generally of high quality and highaffinity, a library has to essentially be set up (animals immu-nized ahead of time) for each antigen. Large naıve librariesoffer the advantage of being able to produce many antibodiesto an assortment of antigens but have in general loweraffinity. When additional diversification is introduced intonaıve libraries in the complementarity determining region(CDR) regions, they are called semisynthetic (5). The librariesare in general made from the IgM mRNA but can be madefrom IgG by using alternative primer sets in the library con-struction stage (Table 1). Synthetic libraries are made usingmodular consensus scaffolds (frameworks) with the CDR anti-gen contact domains encoded by random synthetic oligonu-cleotides (6).

We have divided this chapter into four sections. In Sec. I,we review the development of antibody libraries. We also pro-vide a general classification scheme to encompass all immunelibraries derived from sensitized B lymphocytes for exposed,immunized, and infected hosts. In Sec. II, we review the gen-eral aspects common to all immune libraries. This section willalso emphasize aspects of the strategy to help scientists deter-mine if they should make immune libraries or alternativelymake informed decisions about using another method to pro-duce a particular MAb. There are far too many examples ofimmune libraries for us to review them all within the scope

Table 1 Antibody Libraries

Library V-gene source Diversity

Immune Activated B cells (IgG) Matured in vivoNaıve Resting B cells (IgM) ImmaturedSynthetic Cloned Vgenes Synthetic

Antibody Libraries from Immunized Repertoires 531

of this chapter. However, in Sec. III, we review some recentkey examples of the use of immune libraries to derive MAbsfrom various animal species as well as humans. We highlightpoints upon which our experience with immune antibodylibraries gives practical information. These studies haveexpanded the utility of antibody libraries and show thatlibraries can be derived from the immune repertoire of vir-tually any species for the selection of MAbs. In Sec. IV, wepredict that the use of immune libraries will continue toexpand greatly for other animal species especially with thedevelopment of xenogenic mice. Overall, this review sum-marizes progress in antibody libraries generated fromimmune repertoires.

I.A. History of Phage Display Development

The originator of filamentous phage display was Smith (7).While developing (f-phage) vectors as a biological ‘‘way-sta-tion’’ for cloned DNA, Smith made an important leap. He dis-played a solvent exposed peptide upon the surface of a(f-phage) by creating a fusion protein with a native phage pIIIprotein. This linkage of genotype and phenotype has, nearly20 years later, developed into an entire new field of phage-based ligand selection systems. Smith went on to displaylarge polypeptides (8), but found that they significantlyreduced the phage viability and the phage vector was biologi-cally inefficient compared to today’s phagemid vectors. Thefirst selectable phage display libraries were publishedin 1990 using peptide ligands expressed upon the surface off-phage (9–11).

The ability of prokaryotes to express functional antibodybinding domains of eukaryotic origin was critical to the devel-opment of antibody phage display. The simultaneous publi-cation of prokaryotic expressed variable fragments (Fv) (nolinker) (12) and significantly more stable, and larger, antigen-binding fragments (Fab) (13) demonstrated that Escherichiacoli was capable of expressing functional antibody-bindingdomains, and moreover that some functionality was stillpresent in these molecules. The next major development was


a procedure to clone, en masse, the B cell pool of expressed Vgenes. This came about from the efforts of several independentgroups. Studies in the late 80s and early 90s by James Lar-rick’s lab demonstrated the power of using polymerase chainreaction (PCR) for cloning unknown V genes from hybrido-mas (14–16). Using oligonucleotide primers with designeddegeneracy in the leader regions, or consensus primers, incombination with isotype-specific back primers to the constantdomains, the specific Ig cDNA could be amplified and cloned.While the oligonucleotide PCR-based approaches work wellwith hybridoma-derived RNA encoding MAbs, it was not clearat the time if it would work for the amplification of a library ofpolyclonal antibody responses produced in an immuneresponse. What followed next was the extension of this techni-que to pools of Ig RNA. These efforts were driven largely byincreasing confidence in PCR that was progressing rapidlyat the time. It was realized that a representative polyclonalV gene cDNA pool from an immunized host will inherentlyreflect the increased representation of antigen-specific Vgenes due to clonal expansion and plasmablast formation.Thus, regardless of the inherent bias in the PCR itself, a largepool ofV gene cDNA is representative of the immune repertoireof an immune host and is rich in antigen-specific B cells RNA.

The first antibody libraries were from immune sourcesand were cloned into lambda phage vectors. Lambda phagevectors were used to set up combinatorial antibody libraries,which required screening via exhaustive plaque lift screeningassays. Initially immune libraries from mice and humanswere assembled in this fashion. The expressed V gene cDNApool encoding the Heavy and Light chain antibody-bindingdomains was assembled and cloned separately as distinctPCR amplified cDNA ‘‘cassettes’’ encoding the VH and VL

regions as a Fab fragment (17–19). This marked the begin-ning of a series of very rapid advances in antibody displaylibraries.

The next major advance was the ability to select bindingclones from antibody libraries. The development of antibodylibraries originated mainly from the competing interests oftwo labs. The initial work in immune antibody libraries done


in the United States was by the Scripps Research Institute inLa Jolla, California by Drs. Burton, Barbas and Lerner. Theother group was in the United Kingdom and led by Dr. Winterand Dr. Milstein at Cambridge. These two groups developedoriginally different approaches for obtaining libraries ofantibody-binding domains, scFvs, or Fabs displayed on phage.Those important new advances to the display systems include:scFv display on pIII (20); the first immune mouse scFv library(21); Fab display on pVIII (22); and Fab display on pIII (23);and the first immune human Fab library (24). Soon there-after, naıve human scFv (25) and fully synthetic human Fablibraries (26) were published.

Immunoglobulin mRNA from a mouse immunized with ahapten was used to produce a lambda phage antibody libraryand was screened (not selected) for expression of Fabs pro-duced by the random combinatorial assortment of heavy andlight chain fragments (17). The development of recombinantantibody phage display technology and combinatorial immunelibraries soon followed (27,28). Display libraries brought aboutselection of binding clones, whereby specific MAbs are selectedin vitro upon antigen. Similarly, in the first monovalent com-binatorial antibody display vectors, the VL and VH cDNAswere amplified with primer sets bearing 4 different restrictionsites corresponding to unique restriction sites in the phagemidvector, with two unique cutters each for a total of four enzymes(24). The ability to link genotype and phenotype lays in thegenetic fusion of the heavy chain region to the gene encodingthe C-terminal domain of the minor phage coat protein pIII.This in-frame fusion protein, in the Ecoli coinfected withhelper phage, is coloaded onto the surface of phagemid parti-cles containing the plasmids with the f-phage packaging sig-nals. This ushered in the era of phagemid biopanning.

Phenotypic mixing in an E. coli host coinfected withhelper phage resulted in the assembly of fully infectiousphage particles. The particles carry the combinatoriallibraries of displayed Fab fragments linked to the V genes,which encode them. The V gene cDNAs used in these earlylibraries were cloned sequentially with two independentrounds of ligation into the phagemid vectors. Generally the


VL cDNAs were cloned in first, and then the VH cDNAssecond. The oligonucleotide primers were designed to haverestriction sites to match either the heavy or the light chain‘‘site’’ in the vector using two unique restriction sites eachto achieve directional cloning (29).

Today immune phage antibody libraries can select MAbsfrom virtually any species with enough information knownabout the V genes. This is because the methods are foundedupon an understanding of the molecular genetics of the spe-cies under study, rather than the availability of stable mye-loma cell lines or methods of cell immortalization. Thephage antibody technique has been used to generate MAbsfrom a wide spectrum of species. In this review, we confineourselves to phage-based display systems although severalother systems are available, including yeast (30), ribosome(31), and bacterial display (32).

I.B. Classification of Immune Libraries

Immune antibody libraries are created from the antigen-sensitized Ig repertoire of the host’s available B lymphocytes.A sensitized host is required to produce an immune library.The technique is dependent upon the ability to recover theexpressed Ig repertoire from recoverable B-lymphocytemRNA, and the construction of a representative antibodylibrary displaying functional antibody fragments for the selec-tion of individual clones against an antigen source. We define‘‘Immune libraries’’ as antibody cDNA expression librariesfrom the available B cell lymphocyte pool, wherever its origin,of any host, which has been immunized, infected, or exposedto an antigen (endo or exo origin). This definition includesexperimentally immunized animals, any species havingexperienced an infectious disease, exposure to a pathogen,toxin or venom, neoantigen exposure through the develop-ment of cancer, breakdown in tolerance in autoimmuneresponses, or any other example where an antibody responsehas occurred. Clearly immune libraries exist as a spectrum ofimmune states with differing antigen reactivity dependingupon the B cell source and multiple factors including proper-ties of both the host and the immunogen.


Immune libraries usually represent the IgG subclass of Bcells of an immune animal (33). All immune antibody librarieshave several important characteristics, these include: (1)immune libraries are enriched in antigen-specific antibodies;(2) immune libraries include affinity-matured-bindingdomains when derived from species that undergo this anti-gen-dependent process. Clonal selection leads to the enrich-ment of antigen-specific B cells, and somatic mechanismssuch as hypermutation and receptor editing followed by selec-tion lead to affinity maturation. These processes are discussedfurther below.

Figure 1 (Caption on facing page)


Immune libraries are inherently biased for binding clonesagainst the immunogen. The in vivo immune system, whichhas evolved over millions of years, is still the most efficientmeans to produce high-affinity antibody to a foreign antigensimply by injecting it into a host and allowing a normalimmune response to proceed with its natural molecular andphysiological mechanisms. Additionally, it is of great interestand importance biologically to clone the heavy and light chaindomains of antibody responses naturally produced in vivo togiven antigens and pathogens.

We have implemented an taxonomical classification ofimmune antibody libraries. Whereas in the past antibodylibraries were classified based upon display fragment formatsor antigen selection strategies, the rapidly expanding field ofimmune antibody libraries no longer permits this. We haveclassified immune libraries based upon the species of origin(Fig. 1). Figure 1 depicts all of the species in which antibodylibraries have been published to date as well as the potentiallibraries that may be constructed from immunized fish species.

Figure 1 (Facing page) Taxonomical classification of antibodylibraries. This figure depicts the main animal species from whichimmune antibody libraries have been used to select monoclonalantibodies. Laboratory animals and humans have been used morefrequently due to earlier efforts to understand the antibody genet-ics. Libraries from large animals (including cattle, sheep, camelids,horses, pigs, and dogs) are not far off. Human immune libraries canbe broken down into the types of disease. In many cases, a specificantibody is sought for protective capacity, to prove or imply role of aprotein in pathogenesis of disease or to target a tumor or other self-proteins. Nonhuman primates are also used because of their closerelationship to humans. Mabs from primates should prove to beinterchangeable for human therapy and are also of great interest.Primates have been used to perform ‘‘mixed’’ immunization experi-ments with hepatitis and have revealed that multiple antigens canbe used to make immune libraries for selection of Mabs to multipleantigens. Fish represent a hypothetical immune antibody librarysystem where the genes are known and scaffold has been madebut no publications on immunized fish antibody libraries to derivefish Mabs have yet appeared in the literature.


Monoclonal antibodies have been selected from antibodylibraries produced from immune repertoires of laboratoryanimals (mice, rabbits, and chickens); large animals (sheep,cattle, and camelids); nonhuman primates (macaque and chim-panzee); and humans. This species-specific classification is bro-ken down further for humans and it is based upon the type ofdisease under investigation. Several other species, includingaxotol (34), rat (35), horse (36), dogs (37,38), cats (39), mink(40), and swine (41), have enough molecular informationknown to develop reasonable antibody libraries, butMAb selec-tion from these types of libraries remains to be demonstrated.

Libraries from chimeric systems are considered to bederived from the species, which donated the Ig genes. Theuse of chimeric animals presents a unique immune repertoireand allows experimentation which otherwise could not be per-formed. For example, human antibody libraries have beenderived from SCID mice engrafted with human peripheralblood lymphocyte (human-PBL-SCID) (42,43). In this case,the B cells are fully human and carry the repertoire fromthe individual who was derived from. However the systemallows boosting of the B cell compartment to a specific anti-gen. This is thought by some investigators to be capable ofenriching the human B cell compartment in these huPBL-SCID mice with antigen-specific B cells. Similarly, mice-madetransgenic for human Ig gene loci is another form of chimera(375). In this case, the Ig genes, which are human, are trans-ferred rather than B cells. Thus in V gene transgenic mice, Iggene rearrangement, antibody expression, and B cell toler-ance are all carried out in the context of the mouse system.In the future, when human antibody libraries are derivedfrom these transgenic ‘‘xenomice,’’ they will also be consideredto be of human origin and the primers will need to match thespecies of the donor of the V genes.

II. IMMUNE ANTIBODY LIBRARYCONSTRUCTION

The immune system evolved to produce a diverse spectrumof high-affinity antibodies to foreign antigens. It does this


while accommodating the individuality of each host. Thuseach MAb discovery method has inherent biases, which affectthe repertoire of MAbs, which can be obtained from immuneanimals. Biases in phage display include those inherent toPCR amplification, restriction digestion, bacterial expression,and folding=toxicity, which can affect binding to the originalantigen as a recombinant form. Obviously, it may be very dif-ficult to select anti-E.Coli lipopolysaccharide MAbs using f-phage display.

The development of phage antibody libraries from the Blymphocytes of antigen-sensitized hosts represents a majoradvance in recombinant antibody technology. The ability toselectphage-borne ligandswitha linkedgenotypeaddsapower-ful tool for antibody discovery. In all areas of biotechnology andbiopharmaceutical development, there is a high demand formore efficient generation of MAbs. This section considers somegeneral points regarding antibody libraries. In particular, weemphasize the appropriate choice of immune or other libraries,host immune status, species, and tissue source.

II.A. Monoclonal Antibody Technology

Antibody libraries, at best, reflect the immune status of thehost B cell repertoire from which they are derived. Theimmune status of the host directly impacts upon the successof retrieving monoclonal antibodies from the antibody reper-toire. In any case, the RT-PCR-based cloning procedure cap-tures diversity from the antigen-sensitized available B cellrepertoire. For example, the repertoire in the newborn mouseis restricted by the absence of N-region diversification and thepreferential recombination of certain V gene elements (44). Incontrast, adult mice have developed long-term and highlyreactive memory B cells within the B cell repertoire. Thesesomatic changes are intimately linked to Ig gene assemblyand diversification processes. These assembly and diversifica-tion mechanisms are remarkably conserved among mostjawed vertebrate species. Subtle differences in germlineimmune repertoires likely arose as a result of evolutionaryselection pressure over time (45).


While the actual number and chromosomal location of Iggenes differ from species to species, the elements involved inthe construction of Ig repertoires are highly similar. Thus notonly is the available repertoire shaped in a different back-ground of self-proteins between species, but the germlinearmament itself is different from species to species and evenwithin a species.

Each method used to develop MAbs from immune ani-mals captures a slightly different ‘‘snapshot’’ of the immunerepertoire and the biases inherent in each method preventany one method from representing the entire available reper-toire. However, all methods of MAb development fromimmune repertoires require immunization or exposure of ananimal to an antigen. This sensitization results in a skewingof the antigen specificity of the host’s B cell pool toward the for-eign antigens through the development of immune responses.

II.A.1. Methodological Options for MonoclonalDevelopment

Monoclonal antibodies have been and continue to be routinelyproduced by commercial companies using the hybridoma tech-nique for research and diagnostic tools (377). This is becausethe hybridoma procedure is quite robust for rodents and istraditionally the most efficient means of producing monoclo-nal antibodies to date. However, despite recent advances inthe establishment of new myeloma partners for various spe-cies (1), hybridomas are not as reliable for producing nonro-dent MAbs. The use of immune libraries to select MAbs is aspecialized use of phage libraries which require knowledgeof immunology and the molecular genetics of antibody geneexpression and f-phage display cloning which collectivelyhas contributed to a less widespread use of immune phagelibraries compared to hybridoma production. The latter issimpler, requiring only knowledge of immunology and theavailability of mammalian tissue culture facilities.

The method used to produce MAbs is important and canaffect the type of MAbs discovered. In what has historicallybeen a controversial issue, it is now clear that the identical


monoclonal antibody can be isolated to the same antigen byusing either main technique or an adaptation of them suchas ribosome display. However, this may be a rare find as eachsystem captures a distinct representative cross-section of theB cell response (18,46–48). We would like to point out thatthese techniques are highly complementary and that neitheris likely to provide an exhaustive sampling of the immuneresponse (49,50).

Hybridomas themselves represent another ‘‘immunepool’ to exploit using repertoire cloning. Many laboratorieshave valuable hybridomas to important targets, but no abilityto create recombinant versions of the MAbs or to even cloneand sequence the Ig V genes. The PCR cloning of V genesfrom hybridomas is how many scientists become interestedin recombinant antibody technology. Many excellent papershave been produced on the optimization and PCR cloning ofV genes from human and murine hybridomas using PCRand are a good source for primer sequences (14,15,51–54).More recently, improved methods for the PCR cloning andin vitro recombinant expression of murine scFv (50) and Fabs(23,55) or methods for both (56) have been developed and pro-vide an excellent resource for cloning V genes from hybrido-mas as well as for setting up immune libraries from murine,chicken, human, and primate B cells. A good example of thecomplementary nature of libraries and hybridomas was theproduction of a Fab library from antigen-specific hybridomasof a mouse immunized with a chemical hapten (57). Theydeveloped a method that allowed for the direct chemical selec-tion for catalysis from antibody libraries, wherein the positiveaspects of hybridoma technology were preserved and weresimply reformatted in the f-phage system to allow direct selec-tion of catalysis in vitro. Through this chain shufflingapproach, they identified novel catalytic reactivates thatwould likely not be available through the direct screeningapproach. This two-step procedure based upon an initialhybridoma screen, followed by catalytic selection withphage-antibody, is clearly cumbersome and requires expertisein both phage display and hybridoma development. However,it is an incremental advance toward new approaches, which


may in the future combine the initial selection and screen forcatalytic antibodies in a single step.

V gene cloning preserves the specificity of a hybridomaclone. The cloning and sequencing of the V genes from ahybridoma are increasingly important as it further charac-terizes the binding domain for proprietary purposes as eachbinding domain has unique identity inherent in the V genesthemselves. Laboratories with programs in both recombinantantibody technology and hybridoma development are at a dis-tinct advantage as these technologies support each other verywell for patent protein.



The choice of method used to produce MAbs dependsupon a large number of factors. This even includes the down-stream purpose for making the MAb. Figure 2 outlines thegeneral flow of producing MAbs from Immune libraries com-pared to hybridoma production followed by recombinant clon-ing. The important thing to keep in mind is that the type oftests used to screen=select for a given MAb should be as closeas possible to the intended use for the MAb. For example, ifyou need to develop MAbs for a diagnostic competition ELISA(C-ELISA) then either the first or second level of screeningshould be in the format of the C-ELISA. This is provided that

Figure 2 (Facing page) MAb generation exemplified by phage-dis-play and hybridoma technologies. These two technologies are usedto produce MAbs for research, diagnostics, and clinical develop-ment. On the left is depicted the procedure for obtaining MAbs fromantibody libraries using phage display. Depicted is Fab display butthird-generation vectors like pComb3X are capable of displayingscFv as well. Binding phage is enriched through successive roundsof selection on antigen. The best MAb fragments are identifiedthrough screening of the eluted phage. The right side shows the pro-cedure for generating MAbs from hybridoma technology. Immunesplenocytes are immortalized through fusion with myelomas anddrug selection. Clones producing binding MAbs are identifiedthrough screening for binding antigen. While an IgG is depicted,other isotype such as IgA or IgM could also be identified. The path-way is identical and only secondary reagents need be adjusted forgenetically engineered chimeric mice or normal mice. The middlepart shows the procedures for taking MAbs produced in either path-way into the clinic. MAbs produced from nonprimates in eitherpathway need to be genetically engineered to be more human insequence. These modified MAbs must be tested for safety, and mod-ified in vitro for reduced immunogenicity and reconstructed foroptimal full-length Ig expression. Any time an in vitro modificationis performed, the new version of the MAb molecule must berechecked for binding and safety characteristics. The current indus-trial practice, regardless of the technology used, is to clone the finalcandidate MAb into mammalian expression vectors and to establishstable cell lines for high-level expression through gene-amplifica-tion or other procedures. Ag, antigen.


antigen is optimized and used to identify binders in primaryscreen in a viable ELISA. ELISA conditions must be withinrange of MAS sys 50–400 mg=mL.

Several characteristics of repertoire cloning and immunelibraries will continue to make phage display an advanta-geous technique relative to hybridoma fusion methods. Theseinclude: (1) the ability to select isotype-specific libraries(including IgA (58) or IgE (59)) or to rapidly convert selectedbinding fragments to full-length antibodies of any human orother animal’s isotype in vitro (60); (2) the ability to affinityselect binding clones from a library and not merely screen;(3) the ability to simply freeze down libraries as cDNA andscreen again in the future against other antigens (there is onlyone shot at screening with hybridomas); (4) the ability to nega-tively subtract background phage allow antibodies to com-plexes of antigens to be selected; (5) phage display lendsitself to all molecular approaches as you can immediatelysequence the clones based on vector-specific primers, andpursue any in vitro modifications promptly; (6) the potentialfor high-throughput screening of antibody libraries (61). Therandomization of VH and VL pairs which occurs inherentlyin phage display may be considered as a drawback of thetechnique but lends itself to sample an even more diverserepertoire not shaped by in vivo tolerance; (7) the ability toselect nonrodent MAbs. The only prerequisite for immuneantibody libraries is that the Ig genes of the species in ques-tion are known in sufficient detail to allow design of oligonu-cleotide primers for PCR cloning of antibody-binding domainsfrom immune lymphocytes.

Decisions concerning which approach to use must bebased largely upon the expertise of the lab, as well as theend use for the MAbs. For example, when human IL-5 wasused to produce MAbs from both hybridoma and the immunelibrary methods, completely different representative MAbswere discovered (49). However, this is not surprising giventhe vast redundancy of the immune system, the differencesbetween the methods, and the fact that the antigen is con-served. Also, the number of animals used in the study onIL-5 is small and the library sizes are large so it is really


not useful to conclude one method is better than the other, asboth work.

If a purely diagnostic or researchMAb is needed to a givenantigen which has been cloned and available in pure and largeamounts, then the development of MAbs via the classical mur-ine hybridoma is the simplest and most recommendedapproach for themajority of cases. In keeping with the comple-mentary nature of the technologies, hybridoma-derived MAbscan always be converted to recombinant forms as thehybridoma line provides an unlimited source of specific IgmRNA (62). This is particularly useful for previously identi-fied hybridoma-derived Mabs, which have demonstrablepotent biological properties. Furthermore, as Ames et al.(49) point out, by back-shuffling hybridoma-derived neutra-lizing VH genes into the light chain immune libraries, novelVH=VL-neutralizing Fabs were isolated, showing the powerfulcombination of these two methods of MAbs discovery. It isclear that for generating nonrodent-derived MAbs theimmune library is the most reliable method of choice.

II.A.2. Factors Influencing Choice of Category ofAntibody Libraries

There are many factors which impact upon the choice oflibrary format. We recommend the best choice of libraryaccording to the application of the desired MAbs in Table 2.Laboratory animals are perhaps the best choice for producingMAbs for research tools. They are small, easily handled andhoused, and inexpensive. For veterinary diagnostics, largeanimals of the homologous species, as the disease, are optimalfor the highest accuracy in validated tests. For human thera-peutic antibodies, fully human or some primate MAbs arerecommended source of lymphocytes as they require nohumanization, and will have much lower odds of containingcryptic T cell epitopes.

For therapeutic purposes, every MAb produced regard-less of the technology must be tested for pathologicalcontra-indications. Thesewould likely be due to potential T cellepitopes (in synthetic libraries, in mice-bearing human B


Table 2 Factors Influencing the Library Choice

Antibody application

Antigen restriction Library Research Diagnostic Clinic

Available immunogenic nontoxic ImmuneLaboratory animals Best choice Recommended RecommendedLarge animals Recommended Best choice WorstPrimates (human) Worst Worst Best choice

Unavailable nonimmunogenic toxic Naıve synthetic Chapter 20Chapter 21

546

Berry

andPopko

v

cells, and in V gene transgenic mice), alternative glycosyla-tion patterns (all murine-derived hybridomas), or even seren-dipitous autoreactivity possibly generated through somatic orin vitro modifications. Indeed, it is incorrect to imply thatfully synthetic human MAbs have therapeutic potential with-out advising that they will need as rigourous testing as non-human antibody. Humanization (Chapter 14, this book) ordeimmunization (63) offers strategies to avoid, mask, orredirect this human immune surveillance of these molecules.These strategies include the genetic sewing of murine vari-able regions to human constant regions as ‘‘chimeric’’ anti-body molecules, the deimmunization by removal of potentialT cell epitopes, and the humanization of nonhuman antibo-dies by grafting nonhuman (rabbit, mouse) binding residuesonto a human antibody framework (64). These proceuresare in themselves very technical and extremely time consum-ing and should be avoided if possible as they can also lead tothe appearance of modified fine-specificity or unwanted cross-reactivity on their own (3),whereas the natural immune systeminherently, and rapidly, disallows these potentially deleteriousreactivities. Conversely, humans are a poor choice for the pro-duction of antibodies for diagnostic or research purposes, asyou cannot hyperimmunize them and the amounts of B cellsharvested must be within ethical limits. Finally, if an antigenis toxic to all forms of life, we would recommend either the inac-tivation (toxoid) of the molecule if possible, or to use premadevery large synthetic and=or naıve libraries.

Immune libraries can be custom designed to select speci-fic classes of antibody. Scientists mainly seek IgG MAbsas they have excellent neutralizing properties and haveusually undergone affinity maturation and thus have the IgV genes, which encode MAbs, which have been optimizedfor antigen recognition. In general, it takes around 2weeksfor B cells specific for conventional antigens to class switchand undergo affinity maturation (65). This must be taken intoconsideration when trying to derive IgG MAbs regardless ofthe downstream method used to produce the MAbs. For exam-ple, laboratory animals are generally hyperimmunized inorder to produce high-titer IgG antibody responses, which


are largely expressed from affinity-matured B lymphocytes.Antibody of non-IgG classes can also be selected in order tostudy particular diseases or immune compartments. Forexample, human IgA (58) and IgE (59) have been selectedfrom immune libraries to study Respiratory syncytial virus(RSV) and allergy, respectively.

Immune libraries usually provide specific MAbs againstone antigen (immunogen), although there are a few caseswhere libraries have been used to select MAbs to multipleantigens or infectious agents. For example, when animalswere initially coimmunized with multiple immunogens simul-taneously and thus had enriched the B cell pool in antigen-specific B lymphocytes to multiple antigens (66a), or inhumans where the patients were known to be convalescentto multiple infectious diseases (67).

It is possible to reduce the time needed to elicit an IgGantibody response for some antigens. The serum IgG antibodyresponse, which normally takes about 2weeks to develop, hasbeen elicited in a matter of days to peptides, proteins, andwhole virus under specific conditions (68–70, respectively).In the case of peptides and proteins, the antigens were deliv-ered in vivo to CD11cþ-dendritic cells by linking the antigensto hamster antimouse CD11c in an immunotargeting app-roach (71). Similarly, extremely rapid B cell kinetics is char-acteristic of the high-affinity neutralizing antibody responseproduced in mice to vesicular stomatitis virus (VSV). This fastresponse does not incur further affinity maturation, andshows only a subtle shift in the V gene usage as the responsedevelops (72,73). This would serve to reduce the costs of MAbproduction by reducing both the amount of antigen needed toproduce an immune response, and by reducing time needed tohouse animals.

It is difficult to produce antibody responses against endo-genous self-antigens. This is especially true for self-antigenswith high conservation between species. Rodents are phylo-genetically related to humans, and have high identity in manyof the important regulatory molecules and ligands of theimmune system. This can pose problems in terms of immuno-genicity and immune recognition of highly conserved antigens


for which classic tolerance may be observed. In this case,immune libraries from other more distantly related animalssuch as rabbits or birds may be useful as they are more likelyto recognize these antigens as foreign.

Immune libraries are not particularly useful for fishingout MAbs to targets that the host immune systems were notsensitized against. In principle, large prefabricated nonim-mune (naıve) or synthetic antibody libraries are slowly becom-ing a source of recombinant MAbs with increasingly usefulaffinities. These kinds of libraries offer the possibility of selec-tion of high-affinity antibodies to any desired antigen withoutthe need for involving live animals or immunization, or for theinvestigator to be involved in the library construction (3,4).Theoretically, a single library made in a one laboratory couldbe used by many others to select antibodies to diverse targets.

In general the size of the library is proportional to thechances for selecting a good clone. For antibody libraries cre-ated from immune repertoires, only 106 individual bacterialclones are likely an adequate size to ensure selection of ahigh-affinity clone (74). This is in contrast to naıve or syn-thetic antibody libraries, which require enormous diversityon the order of at least 1010 clones to ensure selection of a rarehigh-affinity clones (6,33,75–78). One advantage to thesetypes of libraries from large synthetic repertoires is that a sin-gle library can be used for the selection of antibodies againstany antigen in theory (4), although the affinity of these anti-bodies tends to be lower (79).

The availability of synthetic or nonimmune librariesoffers a potential solution to the problem of identifying anti-gen binders in cases where it is impossible to immunize (lackof antigen, antigen is toxic or not immunogenic). However, weand others (80) feel that the naıve or synthetic libraries maybe inferior to immune libraries. The current success rate inutilizing pre-made naıve and synthetic libraries to isolate bin-ders, as inferred from the literature, is not widespread,although we understand that most of these data have beenobtained by companies and are not published. Only antibodyrecovered from human lymphocytes or chimpanzees, whichare essentially genomically identical to humans, and which


is expressed in a human cell line, should be most suitable foruse in humans. All others (naıve, synthetic, chimerized,humanized, transgenic mice, etc.) could fail in clinical trialsbecause they are targeted by the host, containing new T cellepitopes, improper glycoslyation, and improper folding. Theycould also be dangerously crossreactive with epitope onhuman tissues. Moreover, since the MAbs obtainedfrom these nonimmune or synthetic libraries are not maturedagainst the antigen of interest, we expect that the affinitieswill be low so that extensive in vitro affinity maturation stepsare required to improve on their binding capacity (80). This isas much work as the initial library construction and screeningfor an entire second library. Therefore, in the end, it is likelymuch faster to use the potential of the immune system toenrich the antigen-binding B cells via clonal expansion andto perform the affinity maturation, and then to generate arelative small immune library.

II.A.3. Sources of Blast Cells

There are several important aspects to consider about thesource of the B cells before creating an immune library.In terms of both sources, the species of origin and the tis-sues are the most important. For example, if human thera-peutic antibodies are required then human libraries mayseem to be the most logical source of B cells. However, Bcells may not be readily available from convalescent,immune, or exposed humans. There must be a bona fide Bcell response to the antigen of interest. In other cases, itmay not be ethical to acquire the best tissue sample fromthe best source. For example, it may not be ethical toextract bone marrow from human immunodeficiency virus(HIV-1) resistant or even long-term nonprogressor indivi-duals living in Africa who are already at high risk foracquiring HIV or other debilitating diseases. As an alterna-tive, closely related nonhuman primates such as chimpan-zees or macaques have been used as a source of B cellsfor immune libraries to Hepatitis E and simiar immuno-deficiency virus (SIV) (81,82).


The source of tissue is critical for immune library genera-tion. The number of antibody reactivities is influenced by prop-erties of theantigen, thenumberof antigenexposures, or boostsaswell as inherent properties of the host immune system.How-ever, success dependsuponadequate samplingof the sensitizedB cells. Clearly, these issues raise another question regardingthe stage at which a B cell is most useful for immune librarygeneration. While comprehensive studies have not yet beendone, the plasmablast stage is thought to be the most usefulfor immune library generation as they contain about 100–300timesmore specific Ig mRNA than a resting B cell (83,84). OncetheB cell becomes terminally differentiated, it becomes greatlyenlargened. This is in order to accommodate the increased sizeof the endoplasmic reticulum, increased cellular volume, andhigh volume transcription, which is occurring in the nucleus(Fig. 3). Fully differentiated, antigen-activated blast cells arerapidly present in presensitized animals following a recallresponse as the specific-IgG serum titre is elevated by day 7.In immune mice, it has been estimated that somewherebetween 1000 and 10,000 antigen-specific B cells are generatedin response to a very complex antigen-like whole virus (85).

Tissues rich in antigen-specific plasmablasts, such asbone marrow (86) and spleen (87), are ideal for generatingantibody libraries provided they are harvested at appropriatetimes. Table 3 outlines most of the B-lymphocyte sources,which have been used for laboratory and large animal as wellas for human=primate library production. Clearly, some tis-sue sources are not easily accessed for some species, such asthe bone marrow in mice. Peripheral blood contains a limitednumber of B cells secreting antibody (few blasts) and thus isnot an optimum source of lymphocytes for the generation ofrepresentative immune antibody libraries (88), but it is easilysampled and does not require cadaver or surgical procedures.The spleen of adult mice contains about 54% B cells, whichmake it ideal for immune library generation or hybridomaproduction (89). Of note, the lymph nodes of large animalsand even the cervical lymphocytes collected from swabs ofhumans have served as successful B cell sources for immuneantibody libraries (68,90).


For repertoire cloning, one must try to maximize theantigen-specific Ig messages in the mRNA population. Whilethis is done proactively by using primer sets that amplifythe mRNA of isotypes of heavy and light chains found to bindthe antigen of choice, the timing of harvest can be critical aswell. Lymphocytes harvested from different sites, or different

Figure 3 B cell blast. B cells become activated and differentiate into memory or blast cells. The blast cell becomes greatly enlarged toaccommodate the increased production of antibody. Note the largegolgi and endoplasmic reticulum and the engorged nucleus whichlikely occurs to handle higher transcriptional activity. Blasts areenriched 100–300-fold in the specific Ig mRNA. Each B cell blastproduces a different antibody according to its own rearrangementand somatic modifications. (Reproduced from Kincade PW, GimbleJ. B lymphocytes. In: Paul WE, ed. Fundamental Immunology, withpermission of Lippincott Williams and Wilkins.)


antigens, have different properties, and we would recommendthat investigators anticipate that some optimization of theboost step and the actual day of harvest is required to increasethe likelihood of successful MAb discovery. For example, thereis a delayed response to some proteins expressed in vivo fromviral or naked DNA vectors, and thus the harvest may have tobe extended to 8 or 9days to ensure maximal B cell stimula-tion on the day of tissue collection and recovery of specificmRNA.

In the case of immune libraries derived from laboratoryanimals, the immunization strategy must try to target tissueharvest to coincide with the time of maximal B cell blast for-mation. Hyperstimulation of B cells is routinely used for theproduction of diagnostic antiserum from larger animals andbirds. However, overstimulation of the murine B cell poolwith too many or too frequent repeated injections was foundto produce negative effects upon the yield of antigen-specifichybridomas (91,92). This is likely due to both clonal exhaus-tion and deletion. Prolonged rest following hyperimmuniza-tions, to allow for the recovery of the B cell, has contrastingeffects upon recovery of this pool. Thus immune responsesshould be allowed to wane between boosters to maximizethe effects of affinity maturation in vivo and the peak anti-body titre determined. Thus in general, B cell harvestingshould be done in lab animals about 3–5days or 7–8days post-boost for hybridomas and immune repertoires, respectively.

Table 3 Possible Sources of Blast Cells

Library

IgG mRNA sourcea

PBL Spleen Bone marrowLymphnode

Tissuebiopsiesb

Laboratoryanimals

þ þ þ � �

Largeanimals

þ þ � þ þ

Primates (human) þ � � � þaFor references, please see corresponding species sections.bLive samples of any origins.


II.B. Cloning Strategies

Once an immune source is decided upon, the expressed anti-body gene pool needs to be cloned and expressed in a selectableform. Immune libraries focus upon the amplification of cDNAproduced from the B cells of an antigen-sensitized host. Theuse of the highly specific PCR approach removes the need topurify B lymphocytes from other immune cells. Clean prepara-tions of total RNA are sufficient and all that is required tobuild good libraries. The expressed Ig mRNA pool, the geneticmaterial encoding the antigen-binding regions of Igs, isreverse transcribed into cDNA, amplified by specific primerpairs in PCRs, and ligated en masse into phagemid displayvectors. There is no need to purify mRNA away from totalRNA. The emphasis is on the quality of the RNA. The qualityis greatly affected by the contaminants, which are producedduring harvest and the length of time the RNA is exposed tothe contaminants before being used to generate cDNA. It isadvisable not to delay the library construction at this pointand to generate the cDNA as soon as the RNA is made.

The cloning strategy is usually dictated by the choice ofdisplay system. The choice of vector, the primers, and allnecessary supplies must be on hand and working by the timethe cDNA is prepared. However, before considering the var-ious methods used to assemble antibody libraries from cDNA,one must consider the development of the cDNA amplificationprocess. When RNA is isolated from pools of B cells, there isinherent scrambling of the natural heavy and light chainspairs. This is due to the fact that, upon harvest and lysis ofthe lymphocytes, all of the mRNA is released into a generalpool. These pools are PCR amplified enmasse by heavy or lightchain specific primers and randomly repaired by the assemblyprocesses. It is likely that natural pairs are formed in antibodylibraries and have contributed to the success of this strategy,as clonal expansion will clearly increase the chances of thelibrary being able to recapitulate these pairings (93).

The genomes of most phagemid vectors are relativelysmall and easily purified in the replicative form as adouble-stranded DNA using standard plasmid purification


procedures. In general, the PCR amplified VL cDNA is gelpurified and pooled. The light chain cDNA is cut with the‘‘light chain’’ cloning enzymes, purified again and ligated tothe precut phagemid vectors (24). This ‘‘light chain library’’is transformed into E. coli, amplified by bacterial growth, pur-ified as a plasmid, and cut again with the ‘‘heavy chain’’enzymes. Following this, the VH cDNA are ligated into thevector, which was again transformed into E. coli, amplifiedand double-stranded phagemid DNA once again purified. Thisis the purified immune library, which is used to transformE. coli for packaging into phage particles upon coinfectionwith helper phage. The order of cloning the VL first thenthe VH was believed to be important to avoid the potential lossof VH genes by deleterious internal cutting by the light chainenzymes as the VH genes tend to be more important in deter-mining antigen specificity. This meant that the representa-tive cDNA of the cloned antibody pool was subjected to theeffects of cutting by four restriction endonucleases as wellas the inherent biases of the amplification of the light chainrepertoire in the prokaryotic systems and left room forimproved next generation vectors.

Detailed protocols for cloning most species that havebeen used successfully for the selection of MAbs from immunerepertoires are described in detail elsewhere (29,80,94,95).

II.B.1. Antibody Diversity, Clonal Expansion,and Affinity Maturation

In many animal species, the molecular genetics of Ig genes isnow sufficiently understood to apply the technology ofimmune antibody library cloning for the isolation of species-specific MAbs. Multiple divergent V gene families make upthe primary antibody repertoire of mice and humans (96). Incontrast, other mammals such as the pig or chicken may havea V gene pool comprised of a single main V gene family. Somespecies tend to utilize one light chain class or the other inimmune responses for reasons that are not entirely clear.What is clear, however, is that each species represents aunique pool of antibody diversity (Table 4).


Table 4 Species-Specific Immunoglobulin Gene Diversity and Primer Requirementsa

Total genesb Familiesc Serum LC (%)Nb primer setsd

Species VH VL VH Vk Vl k l (VH=Vk=Vl) Diversity mechanism

Mouse 145 > 97 > 15 4 3 95 5 22 (15=4=3) CJ=SHRabbit > 200 > 4 4? 3 1 90 10 8 (4=3=1) GC=SHChicken 100 25 1 0 1 0 100 2 (1=0=1) GC=SHSheep 10 ? > 1? 3 6 5 95 10 (1=3=6) GC=SHCattle 15 20 > 1? ? 3 2 98 5 (1=1=3) GC=SHCamele 40 n=a 1 n=a n=a n=a n=a 1 (1=0=0) GC=SHPrimate(human)

44 82 7 7 10 60 40 24 (7=7=10) CJ=SH

Abbreviations: LC, light chain; CJ, combinatorial joining; GC, gene conversion; SH, somatic hypermutation.aBased on the data available at www.medicine.uiowa.edu, and www.ncbi.nlm.nih.gov, and Refs. 80, 94, 113, and 200.bTotal number of genes in genome including pseudogenes.cNumber of Ig chain families based upon expressed proteins.dMinimum number for an IgG library to represent every family; for precise numbers of primers used refer to corresponding section.eFor camel single-heavy chains antibodies (HCAb).

556

Berry

andPopko

v

Somatic modifications create huge levels of diversity. Inhumans, despite estimates, which place total possible diver-sity at 1012, at any one given time there are only about 106–107 different B cell specificities comprising the availablerepertoire. Each method used to develop antibody librariesfrom immune animals captures a slightly different ‘‘snapshot’’of the immune repertoire and the biases inherent in eachmethod prevent any one method from representing the entireavailable repertoire. However, all methods of MAb develop-ment from immune repertoires require immunization or expo-sure of an animal to an antigen, which results in a skewing ofthe antigen specificity of the host B cell pool toward the for-eign antigens through the development of immune responses.

Knowledge of the host antibody repertoire is essential forcloning Ig genes and for the generation of immune libraries.In the immune system of most vertebrates, there are severallevels of diversity in the Ig (V genes) usage (97). The first levelof diversity is the fundamental germline diversity. This is theactual number of V gene cassettes, and the individualsequence of the Ig gene cassettes. The next level of diversityis junctional diversity, which includes the combinatorialdiversity and the way minicassettes physically recombine.The variability of the N-terminal domain of V gene was sub-sequently shown to be due to the fact that the N-terminaldomains are assembled from modular genes (VDJ). Thisincludes N (nongermline nucleotides added to the codingminicassette joints mediated by TdT) and P addition (additionof nucleotides at the end of a minicassette which form a palin-drome) (96,98). Occasionally multiple DH regions are insertedor a minicassette is deleted, which can result in further non-germline changes (99). Insertions and deletions in the hyper-variable loops of antibody heavy chains also contribute tomolecular diversity (100). Recently, receptor editing of Igchains, or the conversion of an expressed V gene for anotherupstream V-gene, has been shown to be another source ofantibody diversification (101). Thus in vivo immune responsesserve to generate higher affinity antigen-specific antibodyresponses and a more responsive memory B cell populationin the event the antigen returns to the host.


Despite the general similarities of the immune repertoireof most vertebrates, divergent evolution and shuffling of thereceptors are still ongoing today (102,103). Differences inthe mechanisms used to somatically diversify Ig genes alsofurther differentiate the species. Indeed, the repertoire ofIgs is heterogeneous between all species and can even varybetween members of the same species (102). Indeed, germlineV gene polymorphisms in humans and mice can be attributedto differences in nucleotide sequences among allelic V genes(104–106), D region sequences, and JH region elements(107,108) as well as differences in the absolute numbers ofV genes (109–111). The fact that individuals exist withunique antibody genes supports the notion that the specieshas more diversity than any one individual and that thegermline antibody repertoire continues to evolve (112).

Immunoglobulin (B cell receptor) diversity is generatedin different ways and in alternative primary lymphoid tissuesin different vertebrate taxa. In general, the cumulative use ofmechanisms for antigen-dependent and -independent somaticdiversification of antigen-binding domains correlates inver-sely with the amount of germline combinatorial V gene join-ing. There are two general systems under which mostvertebrates fall for the generation of preimmune B cell diver-sity (113). These are the classical human=mouse diversifica-tion systems, where B cells are generated throughout life inthe bone marrow, and the gut-associated lymphoid tissue(GALT) system, where a single wave of progenitor B cellspopulates secondary tissues early in life and after which theyare not renewed. In mice and humans, a large preimmunediversity is generated by the combinatorial joining of thenumerous VDJ element for VH and VJ for Vk=L (97). In con-trast, most other mammals (including the rabbit, sheep, andperhaps all birds) generate the preimmune repertoire bydiversifying a single or few rearranged V gene segmentsthrough somatic hypermutaion and=or gene conversion inGALT. In birds and sheep, this occurs in what is essentiallyan antigen-independent process, which in birds takes placein the bursa of fabricus (114,115), and for sheep in the gutileal peyer’s patches (116,116a,117). However in rabbits, the


B cells of the GALT require coincident antigenic exposure tothe normal gut microflora during development for optimalpreimmune diversity (118).

Affinity maturation is the selection of B cells bearinghigh-affinity B cell receptor (BCR) in the germinal centerreaction, which leads to a selective clonal expansion, entryinto the memory compartment and the emergence of higheraffinity soluble antibody. Somatic hypermutation (119,120)is the substitution, addition, or deletion of untemplatednucleotides to a rearranged VDJ or VJ segment that occursin the mature B cells of all jawed vertebrates. This is an anti-gen- and helper T cell dependent response in mature B cellsbut also occurs in immature B cells at the lambda locus ofsheep and the rabbit heavy chain in an antigen-independentmanner to generate those species preimmune repertoire(113). Error-prone gene conversion is defined as a homologousrecombination event where upstream V genes are recom-bined into the recombined V gene expression site (121). Inthe antigen-dependent response, each species tends to usesomatic hypermution and=or gene conversion (which mayturn out to be homologus to receptor editing inmice) to furtherdiversify the immune repertoire. When the high-affinity mem-ory B cells are restimulated upon booster, they terminally dif-ferentiate into RNA-rich blast cells.

The reasons for these vastly different approaches to anti-body diversification are not clear. It is clear, however, thatsomatic hypermutation (122,123), class switch (122), andgene conversion (124) (three B cell-specific DNA modificationpathways which are not present in the B cells of all jawedvertebrates) all depend upon the same enzymes to be carriedout. Clearly, this shows that other differences exist to producealternative diversification processes and that much remains tobe learned from studying the V gene diversification processesof various species (113). The diversification mechanisms usedby each respective species are included in Table 4. Rabbits,chicken, sheep, and cattle all appear to have a GALT systemand use somatic gene conversion to mutate their preimmunerepertoire, and somatic hypermutation to make high-affinityvariants during immune responses. Regardless of the


mechanisms of affinitymaturation, many of which differ some-what from species to species, it is the affinity-matured, anti-gen-selected pool of blasts, which dominate immune libraries.

The ability to generateMAbs from a variety of species willbe important not only for veterinary medicine and compara-tive immunology, but also for human therapeutics. Antibodiesfrom different species conceivably target a different array ofepitopes on a given antigen, which is likely going to be veryimportant for human antigens, which are highly conserved(4). Epitopes, which are immunodominant to the B cells ofone species, may not be immunogenic in a different species.While the ability to target functional-binding sites with MAbsmay be of no importance to diagnostic applications, it is criticalfor therapeutic development and biomedical research. Thuseach species produces a novel range of antibody structure toselect potential agonists and=or antagonists. The repertoireof epitopes targeted on a given antigen is not going to be iden-tical between species because the antibody repertoire is differ-ent. Antibodies from nonrodent species have increasing valuein this regard as murine antibodies still dominate the majorityof human disease models and inherent tolerance mechanismsmake it difficult for murine B cells to mount antibodyresponses to antigens conserved between humans and mice.

II.B.2. Primer Design

Repertoire cloning and library selections of MAbs are notbased upon the random immortalization and cloning of theantibody-producing lymphocytes themselves. Instead, oneneeds to know the nucleotide sequences of the expressed Iggenes and to design oligonucleotide primers for RT-PCR-basedcloning. The more diversity in the expressed variable genepool, the more primers are needed to adequately clone theimmune repertoire. However, the cloned Ig pool may reflect apredominantly expressed allele, even if the genetic disversityof a species is quite large. The oligonucleotide primers wereall in general designed from databanks, personal or public,of expressed antibody V gene sequences from hybridomaclones or simply cloned and sequenced fromB cells of the given


species or a related species. For example, murine primers havebeen used to clone rat immune repertoires as they are relatedrodent species and the V gene sequences of the rat were not aswell characterized as those of the mouse (125). Many excellentdatabases exist today on the World Wide Web to assist scien-tists in the alignment and identification of new V genes, Vgene variants, and immunogenetics of variable regions in gen-eral (Immunogenetics database; NCBI Ig Blast).

The highly variable nature of the Ig domains arguedagainst the possibility of en masse cloning of immune anti-body repertoires with any kind of representative cloned B cellpool. However, the original analysis of V-domain variabilityby Wu and Kabat clearly shows that the framework regions(FRs) of antibody variable domains are less variable thanthe antigen-contacting complementarity determining regions(CDRs) (126). Thus PCR primers used for immune libraryconstruction are mainly targeted to the relatively conservedFRs 1 and 4, which also tend to incur fewer somatic muta-tions during affinity maturation. However, highly mutatedV genes, which do incur mutations within the FRs may notbe amplified but would be unknowingly and inherently lostfrom the immune library at the outset. The framework homol-ogy can be grouped between similar genes, which have beengrouped into families according to degree of relatedness fora given host. For example, V genes with greater than 80%identity were designated to be in the same family and geneswith no less than 70% are also included (127). This informa-tion is very important when trying to clone immune librariesfrom hosts with diverse germline repertoires, for example, inmice and man. Mice have a lot of germline sequence variationin the N-terminal regions of their V genes , which poses a dis-advantage as it necessitates the use of many PCR primers tocapture a good representation of the expressed repertoire. Itis expected that the mouse genome project will reveal impor-tant information regarding the de facto germline repertoire ofthe mouse and this will allow more accurate design of murineimmune libraries in the future.

The more complex the expressed repertoire of a speciesis, in general, the more primers are needed to clone the


repertoire in vitro. The primers are generally designed toroughly correspond to a group or family of related V genesbut can cross-prime members of other families due to homol-ogy either of germline or somatically induced. Thus the num-ber of V gene families, and perhaps more accurately thenumber of functional germline VH and VL genes, impactsupon the number of oligonucleotide primers, which need tobe designed and used to clone an immune repertoire (128).Some animals, such as rabbits and chickens, use very few Vgene segments to encode their Igs, which greatly reducesthe number of primers needed to clone the immune repertoire(Table 4). For example, the primer sets need only to focusupon single expressed V genes , to clone the expressedimmune repertoire of other species like the heavy chain Vgene cDNA of chickens. Usually, for most species, a numberof primers corresponding to framework 1 are used in combi-nation with framework 4 or isotype-specific back primers inorder to amplify a library of antibody cDNA correspondingto a single class (for example, IgG).

Species-specific primer sets are generally needed to cloneimmune repertoires. Although the Ig gene families betweenspecies are highly related, they are not exactly the same.For this reason, primers must be designed for each respectivespecies. The PCR amplification and library construction maystill work as it has for highly related species such as mice andrats (125) and humans and nonhuman primates (129). How-ever, the representation of the V genes within the expressedpool is not necessarily the same between species and the pri-mers would have to be evaluated in a case-by-case setting.Thus, the number of consensus primers needed to accuratelyamplify and clone the representative immune repertoire of agiven species roughly reflects the diversity of the germlineantibody repertoire for that species.

In Table 4, we also suggest a minimum number of oligo-nucleotide primers needed in order to generate a representa-tive species-specific IgG library based upon the known ordeduced numbers of V gene families. The actual number ofprimers used could be higher or lower and is discussed inmore detail in Sec. III.


II.B.3. Display Fragments

There are two main fragments used to display the bindingdomains of MAbs. There is the Fab, or the recombinant ver-sion of the fragment responsible for antigen binding, andthe scFv (single chain Fv) (Fig. 4). The Fab contains the heavychain variable region in-frame with the heavy chain first con-stant domain to form what is known as the Fd portion(VHþCH1¼Fd) and whole light chain. In most cases, the C-terminus of the CH1 region is genetically fused in frame tothe C-terminal domain of the gene encoding the f-phage pro-tein pIII, which loads the Fab onto extruding phage.

Figure 4 Representative antibody fragments expressed on thesurface of M13-like phagemid particles. Monoclonal Fab (fragment,antigen binding; FdþLc), scFv (single chain Fv; light chain vari-able region linked to heavy chain variable region), and VHH (camelheavy chain Fv; no light chain and no CH1 region) are representedas pIII fusion proteins expressed on the surface of M13-like filamen-tous phagemid virus particles. These phage-borne proteins link theIg genotype to the respective phenotype (physical binding proper-ties) in a selectable unit. The ability to select phage-ntibodies highlights the key advantage antibody libraries have overclassical hybridoma screening procedures. Full Ig molecules cannotbe selected upon the surface of phage and if desired must be builtfrom these binding domains.


The Fab display systems have several unique features.In Fab libraries, the light chain is encoded as a separate geneand must cofold around the pIII-fused Fd fragment as it isextruded into the bacterial periplasm. It is possible to cosc-reen positive clones produced from in vitro affinity selectionusing light chain detection ELISA in the downstream screen-ing of phagemid colonies, which acts as a suitable control forexpression. This is done simultaneously. By simply coating aduplicate plate with unconjugated anti-light chain polyclonalantibody, and detecting with anti-heavy chain-specificenzyme-labeled antibody (either anti-Fc-isotype or anti-HAtag in the CH-pIII junction of pComb3X vector). This showsif the vector system is actually working and if bacterial super-natants contain any Fab molecules.

There are several configurations of scFv MAbs. The firstexamples of recombinant Fv linked together the two variabledomains using a polypeptide linker (130). Because the linkeris unnatural, many arbitrary configurations have beendesigned for scFvs. In some cases, the light chain variableregion is N-terminal to the heavy chain while in others theheavy chain is N-terminal to the light chain. In general, thelinkers have been based upon flexible glycine residues. Per-haps, the most comprehensive study of the scFv and linkersconfigurations was by Whitlow et al. (131) in which the effectsof linker length on binding affinity and aggregations wereevaluated for an scFv MAb 4-4-20. Linkers may be suitablymodified to incorporate epitope tags and cleavage sites, toremove proteolytic sites, or for optimal binding to an antigen.It is likely that the optimal linker will vary depending uponthe MAb and the purpose, and could be improved for eachclone. Generally, a single linker is used for immune antibodygeneration for a given vector system (29) and is encoded as anoverlap in the oligonucleotide primers used to PCR amplifica-tion, VH, and VL cDNAs. The use of a unique linker sequenceallows forward-and-backward sequencing of VH and VL

inserts at the junction of the two and obviates the need tosequence the entire scFv from the vector sites at the ends ofthe VH and VL. In this way, it is easy to create variationsof the linker once a potent clone is found.


The scFv libraries offer several distinct advantages. ThescFv library is easier to assemble, as it requires one overlapPCR to join the VL and VH instead of two for Fab libraries(one for CH=CL fusion; then for Fd=light chain fusion), whichreduce the number of steps and time needed to construct thelibrary. Other properties of in vitro-modified affinity-selectedscFvs and Fabs are discussed elsewhere in Ref. 128.

There are natural examples of single chain antibodies.SHARKS (113) and camelids (80) naturally express true sin-gle chain antibodies as a part of their humoral response.Immune libraries from camelid species have enabled theselection of in vivo affinity-matured VHH (no CH1) MAbs. Thisis outlined in detail in the large animal section below. Thepoint is that antibody fragments have a demonstrable rolein natural protective immunity and thus recombinant anti-body fragments likewise have potential value in this area(Fig. 4).

Reliable antibody libraries for the selection of intact full-length antibody molecules have not been demonstrated. Thusif whole recombinant MAbs are needed, they must be rebuiltfrom the fragments (Fab or scFv) selected from immunelibraries. In some cases, the properties of the fragments maymake them useful for therapy. For example, the small size ofrecombinant scFvs enables penetration of MAbs into the backof the eye by simply adding the protein in droplets (132). Topi-cally applied scFv was found to penetrate into the anteriorchamber fluid of rabbit eyes in vivo. The engineered frag-ments were stable and resistant to ocular proteases. In somecases, better tumor penetration may be acheived with thesmaller scFvs and this may warrant their use in anticancertreatments (133a). Collectively, these studies show that theremay be a need to produce the fragments themselves, for exam-ple, as a therapeutic modality, and that there are diverse usesfor recombinant antibody fragments.

II.B.4. Display Vectors

The limitations and plasticity of the much larger phagegenomes naturally led to the use of phagemid systems for


the display of libraries of protein domains. The developmentof helper phage with defective packaging signals enabledscientists to focus upon cloning into much smaller plasmids,which encoded antibody-binding domain as pIII or pVIIIfusion proteins. The early discovery of the two functionaldomains of pIII by Smith revealed that the C-terminal portionof the pIII protein (17) was all that was necessary to have pro-teins loaded onto f-phage particles. More recently, other pro-teins have been revisited for the display of proteins(Chapter 2 this book).

The combinatorial approach of cloning antibody domainswas founded upon the successful cloning of the repertoire intothe prokaryotic system. The original experiments, whichdemonstrated that antibody-binding domains could beexpressed and functionally assembled in E. coli, were success-ful because the antibody fragments were secreted from thecytoplasm into the oxidizing environment of the periplasmicspace under the guidance of bacterial leader sequences (27).It is believed that the oxidizing environment and the secre-tory event collectively possibly contributed to the correct for-mation of disulfide bonds and proper folding of antibodydomains. Analogies are obvious between the recombinantexpression of antibody domains in prokaryotic cells and thenatural production of antibody in eukaryotic cells, wherethe heavy and light chains are probably translated as theyare extruded into the lumen of the endoplasmic reticulum.A major caveat is the absence of the full-length Ig moleculesand the eukaryotic post-translational modification machineryin E. coli. This is another bias inherent to the immune librarysystem in that not all antibody-binding domains will be cor-rectly assembled in the prokaryotic periplasm. However, theselection inherently reveals those that are functional.

Many different phage and phagemid systems have beendeveloped for the display of peptides and polypeptides uponthe surfaces of f-phage particles. There are two main displaysystems for f-phage. The first one for peptide display consistsof modified phage genomes with cloning sites in frame withthe major coat protein pVIII for which there are about�2500 copies per particle (134). The pVIII protein is very


small and fusions result in multivalent display, which canaffect the avidity of antibody domains selected. Moreover, lar-ger polypeptides are not well tolerated as fusions to pVIII inphage systems and thus phagemid systems were developedto accommodate expression of two genes for pVIII and therecombinant proteins are displayed via phenotypic mixing(135,136). The multivalent display offered by the major coatprotein pVIII may be useful for selecting MAbs from non-IgG libraries (22).

In some cases, the use of phage instead of phagemids canallow more efficient selection. The naıve human scFv reper-toire was subcloned from the phagemid vector pHEN1 intothe phage vector fdTET to create an antibody library of5� 108 phage clones, which was selected on several recombi-nant proteins (79). In this case, multivalent phage display,compared to monovalent scFv display, resulted in the selec-tion of a larger panel of clones with improved efficiency of dis-play and expression. The average affinity of the clones fromphage libraries was relatively low and also lower than theclones derived from the monovalent phagemid display. Whilethis may be overcome by utilizing further in vitro affinitymaturation steps, this is not simple and clearly shows thatfor the selection of the highest affinity clones monovalent dis-play is still the best solution. However, in many cases affinityis not directly related to biological effect and the increasedefficiency of multivalent phage libraries to identify largepanels of binders after a single round of panning may facili-tate automation.

Phagemids are hybrids of phage and plasmid vectors.They are bacterial plasmids with f-phage packaging signals,the phage origin of replication, appropriate multiple cloningsites designed to genetically fuse the antibody fragment atthe N-terminus of the pIII C-terminal domain as a fusion pro-tein, and a suitable selection marker (usually ampicillin resis-tance). There are about four or five copies of the pIII protein atone terminus of the particle. The pIII is added last to the phageparticle as it comes out of the periplasm and plays a criticalrole in terminating phage particle length (137). Phagemidslack all other structural and nonstructural gene products


required for generating a complete phage. Phagemids can begrown as plasmids or alternatively packaged as recombinantM13 phage with the use of a helper phage that contains aslightly defective origin of replication (such as VCSM13),which supplies, in trans, all the structural proteins requiredfor generating a complete phage (33).

In general, monovalent display of antibody fragmentshas been adopted for use as a phagemid system based uponthe pIII system. For the most part, phage vectors are limitedin the size of the genetic material, which can be carried in thephage chromosome as their genomes are highly plastic. Theuse of helper phage again helps circumvent this drawback.Unlike lytic phage, f-phages and phagemids are replicatedand extruded from the bacterial periplasm leaving the cellunharmed. As the phage=phagemid DNA is extruded, it iscoated with the phage proteins and recombinant versionsare coloaded. Fully infective helper phage with defectivepackaging signals is used to provide all the structural and non-structural phage proteins to package the phagemid DNA. Thecreation of an N-terminal fusion of a peptide or antibodydomain with the C-terminal domain of pIII allows loading ofthe recombinbant molecule upon the surface of phage particles(138). The recombinant pIII-antibody fragments encoded uponthe phagemid DNA are expressed and are coloaded onto theseparticles through phenotypic mixing in the same E. coli hostcell. The phagemid particle displaying antibody moleculeswith the desired specificity and highest affinity can be selectedand enriched in a process known as biopanning.

Antibody libraries generally use phagemid display sys-tems to circumvent potential problems of random deletion ofthe antibody gene inserts from the much larger phage gen-ome. Most phagemid antibody display vectors are about 3.5–5kb in size (Table 5). The large majority of these systemsuse the low copy number pIII (24) system for selection ofhigher affinity antibodies. The original phagemid vector,pComb3, developed by Lerner’s group was designed for Fabdisplay on pIII (24). Similarly, Winter’s group developed thephagemid pHEN systems for Fab and phage fdDOG1 systemsfor scFv display (21,23). Many second and third generation


phagemid display vectors have been developed with improvedon=off systems, and specific constant domains for modularlibrary assembly and cloning (see Refs. in Table 5).

Most commonly used vectors for immune antibody libraryconstruction have evolved from the pCom3 phagemid displayvector (Fig. 5). The third generation phagemid system,pComb3X, assembles Fabs or scFvs by PCR with modularinsertion that is cloned in a single step (56,129). This improvedvector also incoporates epitope and histidine purification tagsfor improved detection and purification, respectively. In somecases, Fab phagemid vectors have been remodeled in orderto express the constant regions of a particular species. For

Table 5 Vectors Most Commonly Used for Immune LibraryConstruction

Principal vectorand derivatives

Displayfragment Display type Reference

pCom3 (24)pCom3H (268)pCom3X Fab=scFv Phagemid pIII (56)pComBov (90)

pHEN1 scFv (23)pHEN4 VHH Phagemid pIII (210)pUR4536 VHH (235)

pSEX81 scFv Phagemid pIII (327)

pCANTAB5E scFv Phagemid pIII UnpublishedpCANTAB6 (329)

pMOC1 Fab Phagemid pIII (330)

pAK100 scFv Phagemid pIII (50)

pSD3 scFv Phagemid pIII (219)pSD3a (66)

pDM1 scFv Phagemid pIII (214)

pComb8 Fab Phagemid VIII (24)

fdDOG1 scFv Phage pIII (21)fUSE5 scFv Phage pIII (331)


example, in order to facilitate expression of bovine Fabs withthe correct structural format the bovine CH1 and Ck domainswere swapped into the pComb3 vector in order to create pCom-Bov (90). There are many other published and unpublishedvectors with similar small but important modifications in pub-lic and private research laboratories. For more information,about phage display vectors see Chapter 2 of this book.

III. IMMUNE ANTIBODY LIBRARY SELECTION

Advances in the understanding and molecular analysis ofexpressed Ig genes have been made for many animal species.While being of no consequence to hybridoma production, thisknowledge directly translates into basis used to generateimmune antibody libraries from mice and humans. Whywould one need to use novel species to generate MAbs? Inaddition to the fundamental knowledge gained from each spe-cies unique immune system, perhaps the most important rea-son to use other species for the development of antibody



Figure 5 (Facing page) A series of pComb3 family vectors forphage display. The pComb3 series of phagemid vectors (56) weredesigned to express recombinant antibody fragments on the surfaceof filamentous phage as fusions with the pIII-loading domain or toexpress them as soluble proteins. In original pComb3 vector, theFd fragment of the heavy chain and the light chain was cloned sepa-rately into two different directional cloning sites. Separate lacZ pro-moters with two ribosome-binding sites (RBS) gave rise to separatepolypeptides that are directed by the same signal peptide to theperiplasm, where they are assembled and displayed on the surfaceof the phage particle as a Fab fragment. Expression of soluble pro-tein required the excision of gene III by restriction digest with SpeIand NheI, followed by re-ligation of the remaining vector. This fea-ture was maintained in all the subsequent pComb3 vector variants.The pComb3H phagemid vector has a different order of the genes(reversed) and a single lacZ promoter which gives rise to a dicistro-nic message containing the light chain and the heavy-chain Fd.However, the presence of two RBS sequences still gives rise to sepa-rate polypeptide chains. The light chain and heavy chain-pIII frag-ments are directed to the periplasm by two different signal peptides.Two asymmetric Sfi I restriction enzyme sites allow single-stepdirectional cloning of Fab or scFv. The pComb3X vector has threeadditional sequence insertions. The amber codon has been insertedbetween the 30 Sfi I restriction site and the 50 end of gene III. Thisallows for soluble protein expression in nonsuppressor strains ofbacteria without excising the gene III fragment. The 6� histidine(H6) tag has been inserted carboxy-terminal to the Fd fragmentfor universal protein purification. The hemagglutinin (HA) decapep-tide tag has been inserted at the 30 end of the H6 tag for universaldetection using an anti-HA antibody. This vector also allows theexpression of both Fab and scFv. The pComBov phagemid vectoris another modified version of the pComb3H designed specificallyfor bovine library construction. The bovine CH1 is produced viaPCR amplications in the usual Fd fragment. However, the bovinelambda constant region (the predominant bovine isotype) has beenpreinserted into the vector in order to silence an internal Sac I site(228). Therefore bovine light chains variable regions need be ampli-fied and cloned without including the lambda constant domain inthe cDNA.


libraries is that they provide a new source of antibodydiversity to tap into.

Genetic and somatic differences exist within the anti-body repertoires of each species. These can be subtle differ-ences or even completely different mechanisms ofdiversification. Thus each species has a different germlineset of V genes, which have evolved to contact antigens overevolutionary time. Many species use different combinationsof mechanisms of somatic diversification and affinity matura-tion. Some species predominantly use a particular V gene or Vgene class. These pools are shaped by the unique developmen-tal patterns, which occur during ontogeny and the generationof self-tolerance in the context of the speciated antigen milieu.Differences such as unique MHC molecules, alternative post-translational modifications, and genetic drift, ensure a collec-tively vast natural antibody diversity for the planet. Someexamples of these differences include the absence of affinitymaturation in the antibody response in the axolotol (34), thelocalized V gene diversification in the peyer’s patches ofsheep (139), the use of mutated versions of a single VH genein the chicken B-lymphocyte pool (140), and an extra Cysbridge in Rabbit kappa light chain (141). Thus each speciesprovides very different opportunities for the in vitro selectionof MAbs to the same target. This is especially true when youconsider that the recognition of foreign antigens by each spe-cies is not necessarily the same. That is the epitopes ‘‘seen’’ asimmunodominant to the immune system of a mouse may notbe dominant to the human immune system. That is alsoanother reason why vaccines must progress through clinicaltrials in humans for efficacy and they sometimes may nottranslate seamlessly between species. Thus novel repertoiresoffer the possibility of targeting alternative epitopes, perhapsnever targeted by rodent or human B cell repertoires againstthe same antigen. Potent MAbs derived from any of these spe-cies can, in turn, be modified in vitro, for example, for higheraffinity or humanization.

Monoclonal antibodies have been derived from antibodylibraries using a variety of selection strategies. These includeselection for binding using purified antigen, selection for binding


on unpurified antigens (cell panning) with and without negativesubtraction, selections for function, selection linked to phageinfectivity, altered elution conditions, competition, bait cap-ture and high-throughput selection and screening. Thesemethods are described in more detail in Chapter 4 (this book)and reviewed by Hoogenboom and Chames (3). In phage dis-play, each round of selection should show some kind ofenrichment of antigen-specific phages over background. Thisis usually determined by whole phage ELISA and byenriched numbers of eluted phages in each successive roundof selection. For detailed protocols, the readers are directed toan excellent laboratory manual on phage display (29).

Once binding clones have been identified, it is useful tooptimize bacterial expression even further. This is usuallydone most efficiently by using a panel of different E. coli non-suppressor hosts to identify the best strain for a given recombi-nant antibody (142). This is done by transforming the plasmidclone into each strain and then testing a small number of indi-vidual supernatants containing the expressed binding domainsin ELISA or flow cytometry. This will allow identification of theoptimal strain for expression of a given antibody.

III.A. Laboratory Animals

Laboratory animals have been used extensively for the pro-duction of immune libraries. This is because the parametersinvolved in generating potent antibody responses are wellknown, and the animals are smaller and easier to handle inclose confines. Monoclonal antibodies have been produced inboth scFv andFab formats utilizing vectors originally designedfor murine=human Ig expression systems.

Mice, rabbits, and birds have received perhaps the mostattention historically for use in MAb production. Mice havebeen very popular due to the production of hybridoma-derivedMAbs. However, there have been many papers published onprimer design and the generation of immune antibodylibraries from mice. The in vitro modifications required tohumanize a recombinant nonhuman MAb, or to alter affinity,or modify specificity of selected nonimmune MAbs are not


trivial. Success is protein-dependent and each MAb’s originalproperties must be reconfirmed in the end product to ensurethat desired properties are maintained. The difficulty isinherent in the fact that nearly each humanized MAb hadto be completed by modifying the humanization selection orscreening strategy.

III.A.1. Mice

We emphasize that for the routine generation of murineMAbs for diagnostic or reagent MAbs especially when anantigen in good supply, one should strongly consider usingthe hybridoma system. This will simplify the screening andproduction enormously, and give you all the MAb you needthrough simple scale-up. Phage display of murine immunerepertoires is mainly used to supplement the hybridoma tech-nology in cases where humanization and affinity maturationare anticipated as downstream modifications for rodent MAbswith therapeutic potential in humans (143,144). In thesecases, generation of clones with an immune library inheren-tlyz facilitates further in vitro modifications.

Murine immune systems have very complex genetics.The murine Ig loci for VH, Vk and Vl are located upon chromo-somes 12, 6, and 22, respectively. While mice and humanshave similar diversification mechanisms and high homologybetween some Ig gene families, the V genes are unique inboth sequence and physical organization on the chromosome.The murine V genes are more clustered and less interdigi-tated than the human V genes. The heavy chain locus islikely the most diverse among species as there are predictedto be more than 100 VH genes (145), 14 DH (146), and 5 JH

(147) genes found on chromosome 12. It has been estimatedthat there are about 140 Vk genes located on chromosome6, that can be organized into about 20 families (148). Thekappa genes, however, are overall less diverse than the VH

genes among themselves (96). These genes have been looselyorganized into families based upon sequence relatedness(127). Information about the mouse genome is growing dailyand can be found at www.informatics.jax.org.


To fully express the murine antibody repertoire, thecomprehensive sets of PCR primers were designed by Orumet al. (55) based on a collection of 90 heavy and 70 light chainmurine IgG sequences from the Kabat and genebank data-bases (55). Two equimolar mixtures of 25 individually synthe-sized primers (one set for the VH and another for VL) weredesigned to represent all known VH and VL subgroups asdefined by Kabat (149). These primer sets, however, do notcover the lambda light chain repertoire, which in mice repre-sents only about 5% of expressed light chains found in serumIgG. This diminished abundance of lambda light chains isassociated with a diminished molecular diversity as thereare fewer germline Vl alleles (96).

There are numerous other studies on primer optimiza-tion for cloning murine immune repertoires. For exampleten ‘‘forward’’ and 10 ‘‘back’’ primers were designed for clon-ing VH and Vk sequences with additional primer pair to coverthe Vl light chain amplification (150). This demonstratesthat as many as 81 separate first round PCR reactions arerequired in order to clone murine immune repertoires intoan scFv repertoire for phage display library construction.Collectively, this highlights two major characteristics of mur-ine antibody libraries produced from immune repertoires.First, given the inherently large number of PCR reactionsthe scFv format, which requires fewer PCR manipulationsby removing the need to amplify CH1 and CL domains, isheavily preferred over Fabs for murine repertoires. Secondly,in most protocols only a representative set of heavy and lightchain variable region primers are used (21).

The first example of phage display of antibody bindingdomains was with a murine heavy chain Fd fragment (20).Filamentous phage carrying V genes that encode bindingactivities was selected directly with antigen, albeit only theheavy chain and not the light chain was present. Indeed, thisshowed that antigen-specific phage-borne Fds bound specifi-cally to antigen and that rare phage (one in a million) couldbe isolated after affinity chromatography.

Winter’s group went on to develop f-phage librariesdisplaying both heavy and light chains of mice (21). Using a


random combinatorial library of the rearranged heavy andkappa light chains from mice immunized with the hapten2-phenyloxazol-5-one (phOx), they selected scFv expressingspecific phage after a single pass over a hapten affinity col-umn. Indeed, fd phage with a range of phOx binding activitieswere detected, and at least one exhibited high affinity (disso-ciation constant, Kd¼ 10�8M). No phOx-binding clones wereselected after two round of panning from unimmunized mice,suggesting that immunization seems to be necessary toensure strong biasing of the B cell pool toward blasts makingantigen-specific antibody for the library construction of therelatively small size (�106 members).

In the 10 years since those initial publications, over 100articles have appeared describing the successful selection ofMAbs from murine antibody libraries from immune reper-toires by phage display. However, only the examples wherethe use of phage display rather than hybridoma technologywas justified in our opinion are discussed below.

Immune display libraries offer the advantage of selectionrather than screening, which is especially valuable for certainrare or poorly immunogenic antigens. For example, a phage-mid display library was successfully used to obtain high-affinity MAbs against Lex (lewis antigen) (151). All of theanti-Lex MAbs of hybridoma origin they previously producedwere of the IgM isotype and have low affinity for antigen(152a). The two MAbs obtained from GM1 ganglioside-immunized mice using phage display have affinities forLex antigen of approximately 1000-fold higher than MAbPM81, which is currently used to purge autologous bonemarrow of leukemic cells before bone marrow reconstitution(153). The higher affinity of the phage-derived IgG antibodyfragments may improve their use in immunodiagnostic andimmunotherapy.

Murine immune libraries but not hybridomas allowed forthe successful selection of prion protein (PrP) -specific MAbs.Antibody libraries were generated from PrP deficient mice,which had been immunized with mouse PrP (154,155). In nor-mal mice, normal regulatory mechanisms of self-tolerance tothe murine PrP restrict the anti-PrP antibody response to


only nonmurine epitopes. Thus the MAbs derived fromnormal mice would not be useful for the development of ananimal model of prion disease in that they would not bindthe mouse PrP proteins. Furthermore, in this case the hybri-domas produced were extremely unstable for reasons that arenot entirely clear. All attempts at producing MAbs via hybri-doma production from PrP deficient mice yielded hybridomacells that failed to secrete anti-PrP antibodies beyond a periodof 48hr. However, the use of phage display as an alternativeapproach resulted in isolation of the novel antimouse PrPantibody Fab28DLPC. Clearly, the technique of phage displayhas in this case circumvented the block in the generation ofanti-PrP MAbs demonstrating that immune libraries providea complementary approach for the production of MAbs. In asimilar approach, an immunized fibroblast-activation protein(FAP) deficient mouse was used for the generation of scFvswith cross-reactivity for both human and mouse FAP (156).A phagemid display strategy was also used to successfullyisolate scFvs against mesothelin after several attempts to pro-duce antimesothelin hybridomas from spleen cells of immu-nized mice were unsuccessful (157).

Reiter and Pastan found that many scFvs made fromhybridoma-derived MAbs could be inherently unstable. Thisinstability is likely due to the fact that these proteins werenot subjected to the biases of prokaryotic expression and dis-play and this may limit their antitumor activity in animalmodels (158). Phage display technology has been used tobypass the hybridoma step in order to isolate scFv MAbs toa mutant EGFR-vIII from immunized mouse spleens (159).The immunotoxin MR1 was created from one of these scFvsand had a higher binding affinity than all of the recombinantimmunotoxins made previously (Kd¼ 11nM). Therefore, itseems likely that standard phage display inherently selectsfor stable scFvs.

Mice have been used to directly generate partiallyhuman scFvs. Recently, Rojas et al. (160) offer a fast and sim-ple way to produce half-human scFv fragments, while keepingthe advantage of being able to immunize animals for high-affi-nity antibodies. Their strategy was to simplify scFv library


construction by using a library made up of a single humanlight chain variable region gene as a general partner formouse immune heavy chains. This light chain gene was con-structed from genomic cassettes of cloned germline A27 geneto the Jk 1 minigene segment, both of which are prominentlyinvolved in human antibody responses. This simplified proto-col thus involved only a single cloning step of VH regions fromRNA of the mouse immunized with human prostate-specificantigen (PSA) into the vector already carrying the invarianthuman VL region. This work clearly demonstrated that it ispossible to select phage-displayed scFvs with high affinity(Kd¼ 3.5 nM for 10A clone) and desired specificity from a rela-tively small (1.5� 105 members) hybrid immune library.

Several examples of antibodies selected from mouse lib-raries are given in Table 6. In summary, the above examplesillustrate several important advantages of using antibodylibraries from immune repertoires for the generation of mur-ine MAbs. First, murine antibodies can be readily selected insome cases where conventional hybridoma technology wasunsuccessful (154,155,157,159). Second, MAbs with higheraffinity against poorly immunogenic antigens can be isolatedby phage display (151). Third, phage display proved to be

Table 6 Mouse

Antigen Library (size) KD, nM Reference

phOX scFv (2� 105) 10 (21)EGFRvIII scFv (8� 106) 22 (159)MUC-1 scFv (> 107) Nd (332)RSV scFv (3.8� 108) 3.6 (232)CD13 scFv (106) 3.3 (333)Amp scFv (107) 1500 (334)CD30 scFv (1.6� 106) 76 (335)FAP scFv (108) 9 (156)PSA scFv (1.5� 105) 3.5 (160)

Antigen abbreviations. phox, 2-phenyloxazol-5-one; EGFRVlll, epidermal growth fac-tor mutant receptor vlll; MUC-1, MUC-1 mucin; RSV, respiratory syncytical virus;CD13, human CD13, aminopeptidase N; Amp, ampicillin; CD30 human CD30 anti-gen;FAP, fibroblast activation protein; PSA, prostate specific antigen.


right choice for generation of scFvs with cross-reactivity forhuman and mouse antigen (156), which is important for thedevelopment of useful animal models. Finally, chimeric mou-se=human MAbs can be directly derived from immunizedmice by phage display technology (160).

In the future, with the advent of the human V genetransgenic mice, it will be conceivable to use immune anti-body libraries from xenomice as a tool for the library-basedhumanization of already discovered and potent murine MAbsto infectious agents. This is especially the case for the use ofgenetically manipulated mice. We predict that transgenicmice with representative immune systems from other animalswill be developed over time. These would be valuable andmuch easier to manipulate and care for than the larger moresentient higher animals. Mice with gene knock-outs, geneknock-ins, immune deficiencies, and even the human anti-body transgenic V gene mice all have potential value for thegeneration of immune antibody libraries.

III.A.2. Rabbits

Rabbits as a species have been widely used for the productionof polyclonal antibody for diagnostics and research. The wide-spread historical use of rabbit antiserum in many laboratoryand diagnostic tests makes this animal a natural choice forthe generation of MAbs. While MAbs are routinely generatedin mice and rats via hybridoma technology, it has not beenwidely used for rabbits. Recently, a plasmacytoma cell linehas been described as a fusion partner for rabbit B cells toestablish stable hybridomas as a source of antigen-specific rab-bit MAbs (161). However, despite this report, the generation ofrabbit MAbs via hybridoma fusion is not a standard techniqueas compared to murine hybridoma technology. This is mainlybecause of the low stability of rabbit heterohybridoma clones.The development of phage display of antibody fragmentsderived from immune rabbits (162,163) has made rabbits aninteresting alternative source of MAbs.

Compared with the other existing sources of MAbs,immune rabbits have several advantages for use in the


production of immune libraries. Rabbit Ig genes and MAbshave been well studied primarily through the work of theMage and Knight laboratories. First, rabbits are known asexcellent producers of high-affinity polyclonal antibodiesagainst many antigens that are weak immnogens in mice.Second, rabbit antibodies are usually of relatively high affi-nity. Third, because most MAbs are generated in mice andrats, there are relatively few MAbs available that reactagainst mouse or rat antigens. This is particularly impor-tant for the development of therapeutic human antibodiesto self-components, comparative immunology, and otherareas of biopharmaceutical development. These antibodiesmust be evaluated in rodent models and are required torecognize and mediate biological effects through both thehuman antigen and its corresponding mouse homologue.Finally, the limited VH and VL gene usage in the recombi-nation events by rabbit B cells (reviewed in Ref. 164) facil-itates the generation of rabbit MAbs by requiring feweroligonucleotide primers for repertoire cloning. Thus the rab-bit has become recognized as a particularly appropriate ani-mal for the production antibody fragments by phage display(165–167).

The rabbit Ig gene repertoire is well characterized (168).Rabbits are unusual among mammals, although they possesmore than 200 functional VH genes, most of the time, onlythe VH1 gene, the most proximal to the D segment, is usedin VDJ recombination in immature B cells (169). Indeed, morethan 80% of rabbit B cells express the VH1 heavy chain of a1,a2, or a3 allotype specificity (170). This contrasts sharply withthe situation in murine or human B cells. Indeed, murine andhuman B cells express antibody recombined from diversegermline genes, which are grouped into 10 and 6 VH genefamilies, respectively (145,171,172).

Rabbits use multiple light chain genes in the primaryrepertoire in contrast to what occurs at the VH loci (173).Similar to mice, most of the rabbit light chains are encodedby a Ck gene. However, in contrast to other mammals includ-ing mice, the rabbit has two Ck genes, Ck1 and Ck2 (174–176).In normal rabbits, approximately 90% of all light chains are


derived from Ck1, termed the K1 isotype (164). The Ck2 geneencodes the K2 isotype of light chains, which is only expressedin wild type rabbits, Basilea mutant rabbits, and allotype-suppressed rabbits (174,175,177). The lambda light chainscomprise only 5–10% of total serum Igs in the rabbit. How-ever, the organization of Vl and Cl genes appears to be simi-lar to that of other mammals (178). Like the chicken, rabbit Vgenes are further modified in secondary lymphoid tissues bysomatic hypermutation=gene conversion events duringimmune responses (180). More detailed information concern-ing rabbit immunogenetics is available in several recentreviews (141,164).

The practical application of the genetic informationdeveloped on rabbits has allowed for the generation of combi-natorial rabbit antibody libraries displayed upon phage. Theconstruction of antibody libraries derived from rabbit immunerepertoires has resulted in the selection of rabbit monoclonalantibodies (162,163,165). The first published report of the useof antibody libraries from rabbit immune repertoiresdescribed the construction of a phage-displayed rabbit scFvlibrary and the selection of high-affinity monoclonal scFvsagainst the recombinant human leukemia inhibitory factor(rhLIF) (162). One of the selected clones, scFv3-3 exhibiteda Kd of 2.8� 10�8M as measured by surface plasmon reason-ance. This finding opened up the use of rabbit immune reper-toires as a new source of antibody diversity.

Rabbit Fabs have been produced against human type-1plasminogen activator (PAI-1). In this second report on theuse of rabbit immune repertoires, a diverse Fab library wasconstructed from the spleen and bone marrow of a rabbitimmunized with purified human platelet a-granules (163).Several Fabs against PAI-1 were isolated after three roundsof affinity selection. This study demonstrates the power ofphage-selection, as PAI-1 is present only in low amountswithin the platelet a-granule. The a-granule itself representsonly about 0.01% of total platelet protein (181). Clone R012exhibited a Kd of 1� 10�9M and did not recognize severalother members of the serine protease inhibitor superfamily.The ability to sort the rabbit Fab library against individual


a-granule proteins (e.g., PAI-1) suggested that the strategycould be readily adapted for the identification of panels ofrabbit Fabs.

Rabbit PBLs have been used to produce Fabs againstmodel antigen. In the report by Foti et al.(165), the authorsshow the feasibility of using immune PBL rather than spleencells and=or bone marrow for antibody library construction.In this paper, they showed that they could still select Fabswith relatively good affinity. A small library (2� 106 clones)was used to select anti-KLH Fab C1.3IV3 which exibited aKd of 6.4 nM. The authors conclude that peripheral B cellsfrom rabbits appear to be mostly CD5 positive recirculatingB cells which represent the primary immune repertoire inadult rabbits and, therefore, are equally suited as the sourceof mRNA for library construction as spleen cells. Further-more, because PBL and plasma can be prepared simulta-neously, this enabled the authors to follow the immuneresponse of individual rabbits by analyzing plasma for anti-body reactivity. This allowed them to selectively choose, byextrapolation, what they concluded were the best sources ofsensitized B cells from the corresponding stored PBLs formRNA extraction for library construction. A similarapproach was used by Chi et al. (183) to successfully isolaterabbit scFvs, rather than Fabs, against human Reg Iaprotein from a PBL-derived library after only one round ofpanning.

Rabbits are large enough to easily provide sensitized Blymphocytes from tissue sources other than spleen. Hawlishet al. (184) generated rabbit scFv specific for the guinea pigcomplement protein C3 by repertoire cloning from three dif-ferent rabbit immune libraries. The libraries were obtainedfrom spleen, bone marrow, or PBLs of the same animals.The DNA sequence analysis of C3-specific MAb clonesrevealed that the blood-derived (PBL) clone BA8 comprisesa VH� 32 heavy chain. This indicates that VH a-allotypes(185) are not only of theoretical interest for a high degree ofdiversity within the libraries, but they may also be of func-tional relevance in the VDJ recombination events leading toantigen-specific antibodies.


Rabbit scFvs have been selected against multipleantigens from multiply-immunized rabbits. Single chainFvs with subnanomolar affinities against the group of hap-tens including mecoprop, atrazine, simazine, and isopro-turon, were isolated from a phage display library(8.7� 108 clones) derived from the rabbit (66a,). In thiswork, the authors demonstrated that anti-hapten scFvswith affinities in the subnanomolar range could be readilyisolated from an immune phage display library derivedfrom a single rabbit immunized with multiple antigens.This avoids the expensive necessity of constructing separatelibraries de novo for each antigen. The highest affinity MAbexhibited an extremely low Kd value of 6.75� 10�10M. Thegeneral utility of the multiple immunization strategy chal-lenges previously suggested limitations of using antibodylibraries from immune repertoires to obtain high-affinityantibodies (186).

Rabbit monoclonal antibody can be used directly as asource of chimeric human Fabs. To facilitate rabbit MAbhumanization, a chimeric rabbit=human Fab library was con-structed (2� 107) clones where rabbit VL and VH sequenceswere combined with human light chain and CH1 sequences(166). A rabbit was immunized with human A33 antigen,which is a target for immunotherapy of colon cancer. Selectedhigh-affinity Fabs exhibited Kd values as low as 390pM. Incontrast, rabbit antibodies selected by scFv display tend toexhibit lower affinities (162,165). This is likely based on theFab being displayed more or less monovalently at the phagesurface and thus being selected for on the basis of affinityand expression. The selection of scFvs is also influenced byavidity due to their tendency to dimerize or form higher orderaggregates depending upon the length of the linker region(187,188). For the humanization, the authors used a selectionstrategy that combines CDR3 grafting with framework finetuning and found that the resulting humanized antibodiesretained both high specificity and affinity for human A33antigen. In the follow-up article by Steinberger et al. (189),Barbas’ group isolated a human CCR5-specific antibody,ST6, from a phage-displayed antibody fragment library


(5� 107 clones) generated from an immune rabbit. Thepresence of human constant domains, rather than rabbit, atthe C-terminus of the variable regions means that anti-human antibody reagents can be used as another means ofdetection of binding or for purification.

‘‘Recently, Barbas’ group demonstrated that rabbits withmutant bas and wild-type parental b9 allotypes are excellentsources for therapeutic monoclonal antibodies (167). Featuredamong the selected clones with b9 allotype is a rabbit=humanFab that binds with a dissociation constant of 1nM to bothhuman and mouse Tie-2, which could facilitate its evaluationin mouse models of human cancer. That study also revealedthat rabbits exhibit an HCDR3 length distribution more clo-sely related to human antibodies than mouse antibodies(167).’’

Rabbit monoclonal antibody may be useful for intracellu-lar therapy of humans. Recently, Goncalves et al. (190) devel-oped a chimeric rabbit=human Fab library of approx 9� 107

independent clones from a rabbit immunized with HIV-1-encoded Vif protein, which is important for viral replicationand infectivity. A selected Vif-specific Fab was converted intoan scFv, which was expressed intracellularly in the cyto-plasm. Folding of the rabbit scFv in the reducing environmentof the cytoplasm can form functional binding sites as wasdemonstrated by coimmunoprecipitation. Furthermore, thetoxicity of the scFv was low as assessed by cell viability inintrabody-expressing human T lymphocytes. These resultssuggest that gene therapy approaches, which deliver Vifintrabody, may represent a new therapeutic strategy for inhi-biting HIV reverse transcription.

‘‘More recently, rabbit scFvs were used for the designand engineering of a bispecific, tetravalent endoplasmic reti-culum-targeted intradiabody for simultaneous surface deple-tion of two endothelial transmembrane receptors, Tie-2 andvascular endothelial growth factor receptor 2 (189a). The find-ings suggest that simultaneous interference with the VEGFand the Tie-2 receptor pathways results in at least additiveantiangiogenic effects, which may have implications forfuture drug developments.


The antibodies selected from rabbit libraries are sum-marized in Table 7. We have summarized this section withthe following points. First, it was demonstrated that antibo-dies could be successfully selected from relatively small(1.5� 106 clones) rabbit immune libraries (162). Second, thesame immune library can be used to select antibodies againstseveral antigens if the animal was coimmunized with thoseantigens (163). Third, subnanomolar affinity Fabs can beobtained from rabbit immune libraries (166) to ‘‘Fourth, rab-bit scFvs can be successfully expressed as cytoplasmic (190)or endoplasmic reticulum-targeted (189a) intrabodies.’’ Withthe increasing availability of transgenic rabbits, the use ofanimals expressing foreign proteins as endogenous moleculesfrom other animals will help generate MAbs to complexed-antigens. This will greatly facilitate MAb selection fromimmune libraries, which specifically bind to complexed anti-gens and to cryptic epitopes, which are only exposed whenbound to ligands.

III.A.3. Chicken

Birds represent another source of MAbs now becoming recog-nized as a new source of antibody diversity, especially for theproduction of MAbs against mammalian antigens whose

Table 7 Rabbit MAbs from Immune Libraries

Antigen Library (size) Kd, nM Reference

rhLIF scFv (1.5� 106) 28 (162)PAI-1 Fab (2� 107) 1 (163)KLH scFv (2� 106) 6.37 (165)C3 scFv (5� 107) nd (184)Simazine scFv (8.7� 108) 0.67 (66a)A33 Fab (2� 107) 0.39 (166)CCR5 Fab (5� 107) 2.7 (189)Vif Fab (9� 107) nd (190)Tie-2 Fab (109) 1 (167)

Antigen abreviations: rhLIF, recombinant human leukemia inhibitory factor; PAI-1,type-1 plasminogen activator; C3, guinea pig complement protein C3; A33, humancolon cancer A33 antigen; Vif, HIV-1-encoded Vif protein.


structure is highly conserved and therefore render only alimited immune response in mice and rabbits due to immuno-logical tolerance. The chicken, therefore, may be the most use-ful small host for the development of new antibodyspecificities since it is located on a different evolutionarybranch from mammals on the phylogenetic tree (191). Theability to harvest large amounts of chicken IgG (often calledIgY) from eggs makes chicken a desirable target host for thegeneration of large amounts of antibody. For example, thepossibility to produce human xeno-chickens expressinghuman Ig-gene loci could result in large-scale antibody pro-duction against enteric pathogens for food additives.

There are a few reports of hybridoma-derived monoclonalantibodies from immunized chickens (192–194), which corre-spond with the availability of some myelomas for this species(195,196). However, the simplicity of the avian expressedantibody gene repertoire makes it very suitable to createrecombinant antibody libraries.

The naıve B cell repertoire of birds is derived somaticallythrough seemingly random and antigen-independent events.B cell diversity is a result of gene conversion events fromupstream pseudo V-region genes and takes place in the uniquemicroenvironment of the bursa of fabricus of young chickens(197–199). Immature B cells are diversified by intrachromoso-mal gene conversions of a single rearranged functional VH1.There are about 100 upstreamVH genes (116,200) which diver-sify the rearranged VDJ genes until they migrate to secondarylymphoid tissues. A similar mechanism is considered to oper-ate in the light chain, which is composed of one functionalVL and JL with pseudo VL-genes clustering upstream of thefunctional gene (197). The repeated gene conversions of evensimilar V genes result in changes to each of the CDRs andcan alter the length of the CDRs via codon additions and dele-tions. B cell development is thus developmentally regulated inbirds and occurs for a limited time in contrast to the continu-ous development inmice and humans. Furthermore, in birds Bcells with rearranged Ig genes migrate in a single ‘‘wave’’ fromthe bone marrow to the bursa where they then undergo geneconversion (113).


Chickens also use antigen-dependent somatic hypermu-tation in combination with gene conversion to further diver-sify mature B cells during an immune response. Theseprocesses combined with antigen selection leads to affinitymaturation of the chicken B cell response (201). Somaticchanges in general make it very difficult to judge whetherthe origin of some somatic changes is germline templated ornovel as the sequences can be very mutated or very small(202,203). Avian B cell development is covered in more detailin the following review (140).

Chicken antibody libraries are simple to construct.Despite the fact that extensive random gene conversion eventsoccur during bursal B cell development, they tend not to effectthe termini of the V genes as much as the internal stretches(204). Thus the 50 and 30 ends of rearranged V-genes are highlyconserved, which mean a single set of primers for VH and for VL

is sufficient to amplify the expressed V-gene repertoire fromimmune chickens. The feasibility of producing specific phageantibody from chickens was first shown by Davies et al.(205). They generated scFv libraries from isolated bursal lym-phocytes of an 8-week-old naıve chicken providing the founda-tion for antibody library development from chickens. In thesame year, the RNA from chicken hybridoma clones was usedas a substrate for constructing a recombinant antibody libraryfor the selection of antigen-specific recombinant chickenMAbs(206). In this case, the authors initially screened avian hybri-domas from immune chickens injected with a human proteininvolved in the pathogenesis of cystic fibrosis. Next, theycloned the V genes from the antigen-specific hybridomas intoa whole IgG expression system in order to produce chimericchicken=human IgG. This work demonstrated the advantagesto having the chicken as an alternative system to generateMAbs.

Antibody libraries have been derived from avian immunerepertoires. The first demonstration of an antibody libraryfrom an immune chicken repertoire was in 1996 and producedMAbs to a model antigen. Yamanaka et al. (207) used thespleen cells of outbred White Leghorn hyperimmunized withmouse serum albumin (MSA) to construct an immune scFv


library of 1.4� 107 members. All five selected MAbs werehighly specific for MSA and demonstrated different degreesof cross-reactivity to rat serum albumin, but not to humanand bovine serum albumin. Thus, the chicken immune reper-toire can provide simple to construct antibody libraries for theefficient selection of MAbs against murine proteins or otherproteins whose structure is highly conserved in mammalianspecies.

Modular-format vectors have been used to select chickenscFv or Fab MAbs from immune birds. It was not until 4 yearslater that the value of using animals located at phylogeneticdistance became a hot topic. Simultaneously, two more arti-cles were published from Barbas and Silverman’s collabo-rating groups at the Scripps Research Institute and theUniversity of California at San Diego, respectively. Thesearticles described the generation of avian MAb fragments byphage display (56,208). In the first article, Andris-Widhopfet al. optimized methods for constructing chicken Ig phagedisplay libraries in the modular pComb3 system and gener-ated combinatorial antibody libraries from spleen and bonemarrow of Red=Black Cornish Cross chickens immunizedwith fluorescein-BSA. In the same study, they went on todevelop methods to construct scFv, diabody, and Fab formatlibraries all from the same immune chicken. This indicatesthe versatility of phage display in obtaining different typesof chicken MAb fragments from the same animal. The sameVH and=or VL regions were selected from different types oflibraries indicating that the format of the MAb molecule (scFvvs. Fab) had little or no impact with regard to avidity vs. affi-nity in this case. The chicken Fabs library was constructedusing human constant regions, which facilitated detectionwith readily available antihuman secondary reagents. More-over, this vector system is expected to greatly facilitate possi-ble downstream in vitro modification processes.

The evolutionary distance of the chicken immune sys-tem from the human was directly demonstrated in the studyby Silverman’s group (208). In this case, they producedchicken MAbs against the evolutionarily conserved antibodydomains themselves. They immunized Leghorn chickens


with the human clan III Ig (V3-23=clan III Vk2) proteins andselected MAb fragments to these highly conserved mamma-lian antigens. They needed these reagents in order todirectly investigate the expression of highly related antibodyclan-defined sets, and thus they needed to exploit thechicken immune system and the selection power of phagedisplay in order to derive diagnostic MAbs for clan III Ig.Using a specially tailored immunization and selection strat-egy, they selected recombinant avian scFvs specific for theclan III products, including those from the human VH3

family and the analogous murine families. Reactivity withthe representative LJ-26 scFv was completely restricted toclan III Ig and had complete nonreactivity with other clan Iand clan II Ig. They went on to demonstrate the utility ofa novel recombinant serologic reagent for studying the com-position of the B cell compartment and also the consequencesof B cell superantigen exposure. This clearly shows how Iggene diversity of nonmammalian antibodies can be usefulin the development of new reagents for biomedical research.

The antibodies selected to date from immune chickenlibraries are summarized in Table 8. In conclusion, the vastvariety of domestic avian species and the simple immunoge-netics of birds make them an attractive and wonderful newspecies for creating immune libraries for MAb development.The studies with chicken libraries also provided the evidencethat in the suitable host the highly conserved antigen sur-faces can be recognized by the immune system. This alsoallows speculation about the relatedness of proteins to whichthe immune species reacts.

Table 8 Chicken


MSA scFv (1.4� 107) nd (207)FITC scFv (9.6� 107) nd (56)FITC Fab (3.8� 107) nd (56)clan III Ig scFv (8.5� 108) nd (208)

Antigen abbreviations: MSA, mouse serum albumin; FITC, fluorescein.


III.B. Large Farm Animals

Large animals of agricultural importance represent anotherarea where the development of immune antibody librariescan be of great advantage. In many species, the understand-ing of the basis of Ig formation is now sufficient for the appli-cation of antibody phage display and this technology has beensuccessfully applied to animal species including sheep (66),cattle (90), camel (209,210), and llamas (211).

There is a general trend to develop validated diagnostictests for infectious diseases in veterinary medicine whichcan be done in the level 2 laboratory without the need to usebovine MAbs, for example, to pathogens such as foot andmouth disease virus, in order to standardize and verify serolo-gical responses and for the development of validated C-ELI-SAs. The ability to produce recombinant MAbs to infectiouspathogens, which affect our livestock, whether from immunelibraries or derived from hybridoma clones, is expected to ben-efit species of veterinary and economic importance.

III.B.1. Sheep

Sheep polyclonal antibodies are routinely used as laboratoryand veterinary reagents. Sheep also represent a promisingnew system for deriving high-affinity MAbs. In contrast tosheep polyclonal antibody usage, sheep MAb technology isstill in its infancy. The attempts to generate sheep MAb byfusing sheep lymphocytes to a mouse myeloma line were tooinefficient to meet the needs in using these molecules for bio-medical and agricultural purposes (2,212).

Sheep Ig genetics are relatively well studied mainly bythe work of the Reynaud and Weill laboratory. Sheep createa diverse preimmune repertoire through antigen-independentsomatic mechanisms. Genetic studies have shown that sheephave about 10 germline VH-genes, which contribute to thegeneration of preimmune antibody diversity of the sheep(213,214). There is strong evidence to indicate that sheep Bcells are diversified in ileal peyer’s patches (116a,215). Unlikechickens and similar to rabbits, in sheep these additionalgermline VH genes are all functional (not pseudogenes) for


both heavy and light chain rearrangement and expression(113,199,216). Sheep have around 60–90 Vl members thataccount for 75–80% of sheep Ig, light chains (217). Studieson the lambda light chain locus which tend to preferentiallyrearranged only a few of these genes clearly showed thatthe rearranged VL genes become modified by somatic hyper-mutation to generate a diverse preimmune repertoire (116a).Indeed, sheep mainly generate antibody diversity in theirimmunoglobulin repertoire by hypermutating the maturerearranged V genes in postrearrangement diversification(199,218). This is an antigen-independent process, which alsooccurs in sheep B cells in vitro under the appropriate condi-tions. Recent funding, however, demonstrate that combina-torial rearrangement plays a much larger contribution tothe sheep Ig diversity generation than is currently acknowl-edged (218a).

There is evidence for affinity maturation in sheep B cellsduring immune responses. The substantial number ofexpressed sheep heavy and light chain genes indicates thatgene conversion may exist as a possible mechanism of gener-ating antibody diversity (117,214), in addition to somatichypermutation (113). These mutations seem to be targetedspecifically to the CDRs, which support the notion that activeimmune responses provide a beneficial bias in the enrichmentof antigen-specific B cells (114).

The design of primers for the generation of sheep MAbfragments from immune repertoires opens up the use of capr-ine V genes for the generation of MAbs. Until recently sheepwere thought to have a single VH gene family with homologyto human VH4 family, comprised of just 10 germline genes.Thus only three primers were needed to clone the majorityof VH genes (213).

Antibody libraries were used to study antibody diversityin sheep (214). During this study, eight new VH gene familieswere identified with homology to human VH families, whichhad not been previously reported in sheep. These findingsindicate a greater level of germline diversity than anticipatedalthough this did not necessarily imply by itself a larger func-tional gene pool. In total, the sheep VH loci is comprised of at


least nine VH gene families and there is additional evidencefor the usage of JH pseudogenes to encode diverse CDR3s(214). The light chains of sheep are more diverse than theheavy chains, and more primers have been designed to clonesheep VL genes (199,217). This requirement is similar to thatfor rabbit antibody library construction yet is still less thanthat required for cloning of human or mouse VL genes (162).

Sheep scFv MAbs have been produced against modelantigens. In 2000, Li et al. used total RNA isolated from thespleens of sheep immunized with the model antigens humanserum albumin (HSA) and chicken egg conalbumin (CONA)to generate an scFv phage display library (66). The Ig V generepertoires were PCR amplified and used to construct an scFvlibrary in a modified phagemid vector (219). A total of 14different scFvs were isolated and chosen for further charac-terization. The sequences of the MAbs revealed typical ovinecharacteristics and diverse Ig, genes were selected from theimmune library revealing the distinct clones, which weregrouped into three Vl and the one VH family, respectively.The sequence analysis indicates that VDJ recombinationcan contribute significantly to V gene diversity in sheepimmune responses. Due to very low expression levels(< 0.2mg=L), only one HSA binder H3 was affinity purified.The affinity of the H3 scFv produced from the sheep immunelibrary was in the low nanomolar range (Kd¼ 1.83nM). Atotal of only 1.5 grams of immune spleen tissue was used togenerate the sheep immune antibody library. Clearly, thisshows that it is unnecessary to process the whole spleen ofsuch a large animal in order to generate an immune librarycapable of generating antigen-specific MAbs, and this obser-vation may save time and considerable effort. Furthermore,this shows that the entire cellular repertoire does not haveto be recapitulated in vitro and raises the possibility of live(nonmortal) retrieval of spleens from large immune animalsvia splenectomy for MAb generation.

High-affinity sheep scFv MAbs have been producedagainst a herbicide using immune libraries. Splenic mRNAwas prepared from a 10-year-old Welsh breed=Suffolk sheep,which had been hyperimmunized against the hapten atrazine,


conjugated to bovine thyroglobulin. The PCR amplification ofsheep VH and VL cDNAs was achieved using a diverse setof primers representative of the immune repertoire of sheep(214). While the sheep immune library was comprised of1.1� 109 members the library used for selection was preport-edly made up of 8.5� 108 independent clones (220), which isstill a massive immune library. Affinity selection was per-formed using several variations of the immunotube techniqueson atrazine-BSA conjugates as antigen. Eluted phage-antibo-dies were screened first in a phage-ELISA, and later on, solu-ble scFvs were evaluated in a C-ELISA to determine specificityand to gauge affinity. This produced a panel of sheep anti-atra-zine scFv MAbs with high affinity. These high-affinity antibo-dies were highly specific for the atrazine molecule and showedlow cross-reactivity with related molecules in ELISA tests.Two of the selected clones (4D8 and 6C8) exhibited Kd valuesof 0.13 and 0.2 nM, respectively. These monoclonal antibodieswill improve current diagnostic tests, which utilize sheep poly-clonal antibody by lowering background reactivity. Moreover,these scFvs are extremely stable under nonphysiological con-ditions. Indeed, two of the sheep scFv anti-atrazineMAbs havea limit of detection of 1–2 parts per trillion and are well withinthe required EC-legislated limit of 100 parts per trillion.

The antibodies selected from sheep libraries are sum-marized in Table 9. In conclusion, sheep MAbs have been gen-erated by antibody (phage display) libraries from immunerepertoires and can provide specific high-affinity probes forbiorecognition.

III.B.2. Cattle

Immune cattle libraries are another important large animalsystem in which progress has been made. Although hybrido-mas have been derived from cattle (221), the unstable fusionpartners, and the simple V gene expression pattern in cattlemake repertoire cloning an attractive alternative. In cattle,B cells are diversified in both GALT and spleen (222).Therefore, it is expected that cattle engage in multiple Igdiversification mechanisms.


The preferential expression of lambda light chains is a fea-ture common to many domesticated species including cattle(199), and the cattle light chain pool is dominated by a singleVl family (223,224). Like the sheep system, the available VH-gene repertoire of cattle is nearly entirely derived from a singlegene family comprising around 15 nearly identical genes. Thebovine Ig repertoire is diversified by both somatic hypermuta-tions and gene conversion events. Indeed, cloned adult cattlemu-VH-gene transcripts revealed evidence of extensive somatichypermutation and very long CDR3s. The length of CDR3 fromV(D)J rearrangements averaged 21 amino acids, which is largerthan other mammalian CDR3s (225).

Cattle antibody repertoires arise by somatic hypermuta-tion and=or gene conversion,which indicates aGALT-mediateddiversification process (226). Robust somatic diversificationmechanisms and stringent cellular selection must be takingplace in these animals as antibodies derived from them inmostcases have higher affinities (Kd¼ 10�10 to 10�16M) compared tothe average affinity of murine MAbs [Kd� 10�9M (1)]. Therelative simplicity of cattle Ig genetics also makes cloning ofimmune repertoires from these animals relatively simple asfewer primers are needed. Interested readers are invitedto review these references for more information on cattle Iggenetics (94,227).

Table 9 Large Animals MAbs from Immune Libraries


SheepHSA scFv (4.2� 106) 1.83 (66)Atrazine scFv (8.5� 108) 0.13 (220)

CattleGST Fab (2� 107) nd (90)

CamelidsLysozyme VHH (107) 5 (210)AM VHH (5� 106) 3.5 (235)RR6 VHH (nd) 18 (236)

Antigen abbrevations: HSA, human serum albumin; GST, glutation S-trasnferase;AM, porcine pancreatic -amylase; RR6, azo-dye RR6;


The first and only immune library from cattle to date (90)demonstrates that enough information about the bovine Iglocus is at hand to develop immune libraries in vitro. O’Brienet al. performed three rounds of affinity selection of Fablibrary derived from a Simmental calf immunized with aGST-fusion protein and selected 23 anti-GST clones (Table9). However, the expression levels of the recombinant Fabsproduced in vitro by E. coli could not be detected by SDS-PAGE or immunoblotting, suggesting that a very low levelof antibody was being expressed. O’Brien et al. went on in2002 to optimize expression of bovine Fabs G77 and L250using the third generation pCom Bov expression vector withcloned bovine CH1 and Ck1 (228). This time high levels ofFab protein were found in the culture growth medium, whichimplies that a reliable method for the generation of bovineMAbs from immune antibody libraries is now available.

III.B.3. Camelids

The serum of camels and llamas (camelids) contains a uniquetype of antibody devoid of light chains in addition to conven-tional antibodies (229). The heavy chains of these single-heavy chains antibodies (HCAb) have a lower molecularweight due to the absence of the first constant domain, theCH1. Camel serum contains about 75% HCAbs, while llamascontain no more than 45% HCAbs (230). The variable domainof the heavy chain is referred to as VHH to distinguish it fromclassic VH (231). The cloning of VHH in phage display vectorsoffers an attractive alternative to obtain the smallest antigen-binding fragment (MW � 15kDa) compared to scFv with bothVL and VH (MW� 30kDa). The VHHs selected from immu-nized camels or llamas have a number of advantages com-pared to the Fabs and scFvs derived from other animalsbecause only one domain has to be cloned and expressed.

The camel and llama VHH sequences belong to a singlegene family, namely family three, which shows a high degreeof homology with human VH3 (231). It is unlikely that camelspossess other VHH gene families, as a representative data-base of germline VHH sequences revealed the presence of


about 40 different VHH genes that all evolved within the VH3subgroup (232).

Cloning the repertoire of antigen-binding VHH frag-ments from an immunized camel is straightforward. Thesingle-domain nature of the VHH simplifies the effort consid-erably as only one set of PCR primers is required to amplifythe entire in vivo matured VHH repertoire of an immunizedanimal (210). Also, no scrambling of VL and VH pairs occursbecause the VL cloning is not required. However, two distinctmethods were designed to avoid cloning of VH fragments intoa VHH pool. The use of PCR primers that anneal selectively onto the hinge of the HCAb and thus will only amplify the VHHgenes (230). Another method is to use pan-annealing primersto amplify all IgG isotypes followed by separation on an agar-ose gel and recovery of the desired shorter fragment (210).More detailed information concerning the current status ofsingle domain camel antibodies is available in a recent excel-lent review by Muyldermans (80).

High-affinity camel MAbs can be selected from immunecamel VHH repertoires. In the first report (210), two VHHlibraries were constructed from the PBLs of a camel thathad been immunized with two model antigens, tetanus toxoid(TT), and lysozyme. Several VHH isolated from the immu-nized libraries are extremely stable, highly soluble, andhighly specific for the target antigens. The affinity determina-tion of one of the selected antilysozyme clones (cAb-Lys3)yielded a Kd value of 5 nM. Later analysis of this camelVHH clone showed that residues at the tip of the CDR3 loopmimicked the carbohydrate substrate of lysozyme, thus mak-ing it a true enzyme inhibitor (234). This report showed thatimmune camel VHH can be viewed as a route to obtain a newclass of high-affinity antibodies capable of targeting epitopesrarely accessed by conventional VH=VL-based antibodies.

Camel VHH heavy chain antibodies have the ability toinhibit enzyme activity. In 1998, Lauwereys et al. demon-strated that functional HCAbs from camels behave quite differ-ently in comparison with conventional antibodies (236). In thisstudy, they immunized one adult male dromedary with twoenzymes, porcine pancreatic a-amylase (AM) and bovine


erythrocyte carbonic anhydrase (CA), and constructed a VHHlibrary of 5� 106 members from camel PBLs. Four differentinhibitory VHH fragments were selected (two for each enzyme)after three round of panning. The AM-specific clone AM-D9binds the target with a Kd of 3.5 nM and inhibited enzymaticactivity of AM with an IC50 of 10nM, whereas the CA-specificclone CA-06 binds the target with a Kd of 20nM and inhibitedenzymatic activity of CA with an IC50 of 1.5mM. In summary,these findings suggest that the selection of VHH antibody frag-ments from immunized camels is a powerful strategy for theselection of a new type of potent and specific enzyme inhibitor.

Llama VHH MAbs can also be selected from imm-une libraries. The first example of the isolation of llamaanti-hapten-specific VHH fragments was published byFrenken et al. (235). Three young male llamas were immu-nized with three azo-dyes (RR6, RR120, and human preg-nancy hormone chorionic gonadotropin. From theconstructed immune libraries, the authors were able to selectantigen-specific VHH fragment-producing clones by simplecolony screening. For all antigens, the percentage of anti-gen-specific clones was over 5%. This high percentage of bin-ders in the immune pool abrogates the need to use phagemidor other selection=display systems and simple screening willreveal binders. The anti-RR6 fragments were further investi-gated and affinities determined for one of the anti-RR6 frag-ments (R5) was found to be in the low nanomolar(Kd� 18nM). This work demonstrated that the llama couldbe an even more practical source of antigen-specific VHH thanthe camel. The last two examples show once again the abilityof immune libraries to be used to generate MAbs to multipleantigens.

The antibodies selected from camel and llama librariesare summarized in Table 9. In conclusion, the VHHs selectedfrom immune camel and llama repertoires have strikinglyhigh-affinity constants for their target proteins that are com-parable to those of Fabs and scFvs (210,235,236). More impor-tantly, the small size of the VHH fragment and extendedCDR3 loops allow access to antigenic sites not generallyaccessed by conventional antibodies (e.g., enzyme active sites)


(210,236). The unique features of HCAbs may facilitate otherspecific therapeutic applications such as drug delivery vehi-cles and as intrabodies where VHH fragments should performbetter than other antibody formats.

III.C. Humans

Antibodies and their derivatives constitute the largest class ofbiotechnology-derived molecules in clinical trials. Collec-tively, MAbs represent about 25% of therapeutics currentlyin development (237), 30% of biopharmaceuticals in clinicaltrial (63), and have a prospective market value of several bil-lion dollars (77). Antibodies have a long and proven trackrecord for therapy including prevention of hemolytic diseasesof the newborn with anti-rhesus D preparations (238), pre-venting chronic hepatitis B in high risk infants (239), and pre-vention of Argentine hemorrhagic fever (240). These andother examples are discussed in more detail elsewhere byBurton and Barbas (27,302).

All of the 12 MAbs currently approved by USA the Foodand Drug Administration (FDA) contain rodent proteinsequences (Table 10). These antibodies which all bear mur-ine sequences have the potential to elicit pathological com-plications when used in humans. For example, patienttrials, which compared the half-life of murine vs. chimericmouse=human MAbs, revealed that the human immune sys-tem specifically responds to the protein sequences of murineorigin, even in the chimera, and the proteins are treated asforeign proteins by the human immune system (241). Thissensitizes the human immune system against futurerepeated therapy thus reducing the efficacy of a MAb treat-ment. Glycosylation patterns on the MAbs themselves arealtered when produced in other species (including mice,insect cells, and plants) and can affect MAb function as well(242).

Antibody libraries and hybridoma technology are nowbringing fully human therapeutic antibodies to the clinic.Most of the more than one hundred new MAbs in clinicaldevelopment are derived from hybridoma technology from


normal mice, which results in either a fully murine, chimerichuman-mouse, or a humanized antibody molecule. The grow-ing trend is to develop therapeutic MAbs of fully human ori-gin which to date can be derived from either phage displayof human or primate libraries or from hybridoma screening

Table 10 FDA-approved MAbsa

Product name Target IgG formatDiseaseindication Company

OncologyRituxanb CD20 Chimeric NHL Genentech=

IDECHerceptin erbB2 Humanized Breast

cancerGenentech

Mylotarg CD33 Humanized AML Wyeth=Celltech

Avastin VEGF Humanized Colorectalcancer

Genentech

Campath CD52 Humanized CLL Millenium=Ilex

TransplantologyOrthoclone CD3 Murine Transplant

rejectionOrthoBiothech

Zepanax CD25 Humanized Transplantrejection

Centocor

Simulect CD25 Chimeric Transplantrejection

Novartis

AutoimmuneRemicade TNFa Chimeric RA, Crohn’s

diseaseCentocor

Xolair IgE Humanized Allergy Genentech=Novartis

InfectionsSynagis F-protein Humanized RSV MedImmune

CardiologyReoPro aIIbb3=IIIa Chimeric Cardiovascular Centocor=EliLilly

Disease abbreviations: NHL, non-Hodgkin’s lymphoma; AML, acute myeloid leuke-mia; CLL, chronic lymphocytic leukemia; RA, rheumatoid arthritis; RSV, respiratorysyncytial viral disease.aLatest FDA approvals can be found from www.fda.gov.bAlso available as 90Y mAb conjugate (Zevalin).


from transgenic animals (237,243). Significantly, around15% of these new MAbs in the clinic are fully human andwere derived from hybridoma fusions from transgenic mice,or from a human antibody library. Antibody libraries thusremain an important source of human MAbs and have pro-duced about 30% of all the fully human antibodies in theclinic (244). Monoclonal antibody represents the future andfinal improvement of the current gold standard in humanantibody therapy, which is the use of pooled human Igs.

Molecular approaches for the generation of human MAbsoffer several advantages over traditional methods such ashybridoma technology or Epstein Barr virus (EBV) immorta-lization. These traditional methods often can result in a biastoward certain B cell populations and the creation of cell linesthat produce only low levels of antibodies or are unstable(245). Since the phage display method removes any technicallimitations to the production of fully human MAbs, there is ahuge effort to produce fully human MAbs for therapeutic usein humans for the treatment of infectious diseases, autoim-mune disorders, transplant rejections, and cancers. Severalkey examples of the use of immune repertoires for the selec-tion of human MAbs to these diseases are highlighted below.

Humans have a diverse germline Ig gene repertoire.Large-scale sequencing has revealed the entire VH locus ofan individual human being. This has shown that humans havea total of 123 VH genes with 39 functional VH gene segments(246). This also revealed that the total human combinatorialdiversity of the VH locus is much smaller than first anticipatedat about 6000 possible combinations. Humans have a total ofabout 82VL genes making up their light chain germline reper-toire. This is comprised of 46V kappa and 36 lambda genes,which are located on chromosomes 2 and 22, respectively(NCBI database). The human preimmune repertoire is mainlymade up from combinatorial joining (junctional) processes.

The preimmune repertoire is highly diversified during animmune response. The repertoire of antibodies expressed inhuman memory responses is highly selected by antigen.Somatic hypermutation contributes significantly to the shap-ing of the immune repertoire in humans and leads to a shift in


the repertoire of VL genes expressed in naıve vs. memory Bcells (247). The sequencing of the human VH gene locus hasmade it clear that primers can be designed to specificallyamplify the expressed repertoire, based upon the functionalgermline genome. However, it is likely that more than thisminimal number of primers will continue to be used to clonehuman antibody libraries as the nonfunctional VH genespotentially can become functional by gene conversion and=orreceptor editing mechanism. Moreover, there is also a lot ofpolymorphism in the V gene loci (111,248,249), which canalter the repertoire in individuals in particular of certain eth-nic backgrounds (250). It is expected that individuals willhave a slightly different repertoire, which shows the total Igrepertoire in humans as a species remains unknown and con-tinues to evolve (251).

There are many publications on the design and use ofoligonucleotide primers for the amplification of human Igvariable region genes. The minimum number of primersneeded to clone the whole human Ig repertoire, which consistsof the 44 VH and 82 VL genes (Table 5), would be enormousand total about 130 primer sets. However, in general, only arepresentative sublibrary of the immune repertoire is clonedbased upon the phenotype (heavy and light chain class) ofthe desired MAbs (24,252). In the case of the most popular for-mat IgG1kl, a minimum of about 25 primers is enough to clonea representative immune library to select binding clones froman immune human repertoire. In some cases, rather thanamplify IgG1 subclass alone, all four IgG subclasses shouldbe amplified based on the patient serum containing IgG2and IgG4 autoantibodies (253). The use of the popular IgG1klformat for immune library construction should be broadenedto include other isotypes for the study of the in vivo reper-toire, because the use of the IgG1kl format alone imposessevere limitations in the diversity of Igs. For an up-to-datecomprehensive review of primer design, we recommend thereaders consult one of several authorative reviews (129,254).

Although phage display has been used to generatehuman MAbs without immunization, there remains consider-able interest in cloning antibodies from immune individuals


for prophylaxis, therapy, and study of the human humoralresponse. Contrary to experimentally immunized animals,the majority of immune human libraries are from naturallyexposed patients. This includes, but is not limited to, exogen-ous antigens such as bacterial or viral infections, rare casesof vaccination in humans, and exposure to endogenous self-antigens in cancer or autoimmune disorders. All of thesesituations fall under our initial definition of immune libraryin Sec. I.

Mice can be used to directly derive human MAbs in spe-cial cases. Usually antibodies are prepared from a recentlyboosted animal such that the B cell pool reflects ongoingimmune responses (255). In humans, this is restricted asethical constrains generally prevent antigen boosting. How-ever, the use of transgenic mice expressing fully humanantibodies (256,257) or of severe combined immune defi-ciency mice populated with hu-PBL-SCID allows for the res-timulation of B cell antibody responses without theseconstraints (258). The earliest example was the use ofimmune antibody libraries derived from sensitized humanlymphocytes via the hu-PBL SCID mouse (42). The hu-PBL mouse was boosted in vivo with TT to which the humandonor had been immunized over 17 years earlier. The splenicmRNA was used to generate immune human Fab library.TT-specific human Fab fragments were isolated throughthree rounds of panning with apparent binding affinities inthe nanomolar range. Following this, more rapid techniquesof using SCID mice to derive human MAbs were developed.RSV neutralizing human MAbs, with therapeutic potential,was isolated following a single round of stringent biopanning(232). This was done by combining the hu-PBL-SCID mousemodel with an scFv phage display library technique. Theauthors were not only able to bypass the meticulous hybri-doma route but also avoided the inherent skewing of humanantibody responses toward the dominant, nonneutralizingepitopes of RSV. This was noted previously as a confounderfor the generation of potent MAbs to RSV when attemptingto clone virus-neutralizing MAbs directly from humandonors (259).


The use of human Ig transgenic chimeras will also allowfor the simple generation of human MAbs via the antibodylibrary approach. The hybridoma technology has already ledto the development of fully human MAbs against Neisseriameningititis (260), the shiga-toxin (261), the human HIV-1virus (262), cancer (263,264), and autoimmune=inflammatorylammatory diseases (265,266). The use of antibody libraries toderive fully human MAbs from transgenic mice has yet to bedescribed in a scientific publication.

There is a spectrum of immune states from whichimmune libraries have been made from human B cell reper-toires. This spectrum depends upon the clinical pathogenesisof the disease and the level of immune exposure of the host(Fig. 1). It is advantageous to be able to collect an enrichedpool of immune B cells. These tend to be found in the marrowof convalescent patients. The optimal time to collect would befollowing recovery of a patient from an acute infection. Inmost cases, the patient will have a vigorous antibodyresponse and then will go on to clear the infection. Immunelibraries generated from such an individual with a highserum antibody titer would be expected to provide a largenumber of pathogen-specific MAbs by selection upon the inac-tivated agent or predominant antigens from the agent invitro. The next best case for successful selection of MAbs fromimmune libraries is perhaps to use the B cells of long-termnonprogressor, patients chronically infected with viruses, likethe 1 HIV-1. In this case, while the infection is not cleared,the B cells are driven to produce very strong immuneresponses, which typically result in the production of high-titer neutralizing antibody to the homologous infecting viralstrain.

Phage display was successfully used to isolate MAbsfrom individuals with demonstrable serum antibodyresponses to a variety of antigens, including infectiousagents such as HIV-1 (267), self-antigens in autoimmune dis-eases (268), and mutated protein in malignancy (252). Sev-eral examples of the use of immune repertoires for theselection of human MAbs to these diseases are discusedbelow.


III.C.1. Infections

There are a numerous examples of successful MAb selectionsfrom human immune libraries against microbes (269,270). Wepresent several examples of human antibody selection fromimmune libraries, which represent the selection of MAbsagainst some of the worst and most feared infectious scourgeson earth.

The first immune repertoire used to select anti-HIV-1MAbs from humans was prepared from a 31-year-old long-term nonprogressor who had been HIV-1 positive for 6 years(267). The individual had high-titer serum IgG to HIV-1gp120 envelope protein of strain LAI. A Fab library of1� 107 primary clones was made from the RNA isolated frombone marrow samples from this person. Selection on gp120 invitro led to the discovery of a panel of related clones. This ledto the discovery of one of the most potent HIV-1-neutralizing MAb, b12. They went on to improve the affinityof these neutralizing MAb, which resulted in the mostpotent human anti-HIV-1-neutralizing MAb to date (272).The importance of this antibody cannot be overstated forthe development of an active vaccine. This renewed hope forthe quest for an HIV-1 vaccine and that antibodies may actu-ally play a role in limiting infection. Indeed, the holy grailof HIV-1 vaccine design would be to use b12-like MAbs torationally develop immunogens capable of engenderingsimilar protective antibody responses in humans. It is knownthat b12 and human MAbs of similar potency are capable ofpreventing mucosal infection in vivo in the SHIV-macaquemodel (273–275). Along with MAbs to HIV-1 envelope pro-teins, scientists at the Scripps Research Institute have madehuman MAbs to many other important viruses (includingRSV, Cytomegalo virus (CMV), Herpes simplex virus (HSV-1)Varicella zoster virsus (VZV) from the lymphocytes ofinfected or exposed individuals under informed consent. Foran excellent review on much of this work, the readers arereferred to Ref. 27.

The production of human MAbs to highly pathogenicorganisms can sometimes be facilitated through the use of


surrogate organisms. Similarly, the recombinant productionof antigen in some cases reduces the need to use live infec-tious organisms for immunization. However, live organismis still required at some stage in order to obtain enoughnucleic acids for cloning of the important antigens. What hap-pens when the organism is not only lethal, but also the protec-tive antigens are not clearly defined?

The creation of neutralizing human monoclonal antibo-dies to Pox viruses is an excellent illustration of the difficul-ties sometimes encountered in producing MAbs (276). Poxviruses are among the largest of viruses in terms of geneticcomplexity. Vaccinia virus has a genome of about 192kb,which encodes more than 100 polypeptides (277). In thisstudy, a panel of Vaccinia-specific Fabs were selected from acombinatorial phage display library made up of IgG andlight chains from a library prepared from about 2� 107 PBLsof a Vaccinia virus immune donor. Plaque reduction–neutralization tests revealed that six of the Fabs were able toneutralize Vaccinia virus infectivity in vitro. This is the stron-gest evidence to date to suggest that antibodies to Vacciniavirus may be responsible for neutralization and protection toSmallpox (closely related to Vaccinia virus). Furthermore,ELISA studies revealed that 15 of 22 Fabs recovered fromthe library were cross-reactive with the Monkeypox virus, ahighly virulent zoonotic relative of the Vaccinia and Smallpoxviruses. However, clone 14, which had the best Vaccinia virusneutralizing activity, failed to bind to the Monkeypox virus,again revealing the antigenic complexity of these viruses.

There are currently no vaccines or effective treatmentsfor filovirus infections. Viruses such as Ebola and Marburgcause a severe hemorrhagic fever with high mortality inhumans (278). A small percentage of humans infected withthese viruses do survive and may be the important clue todeciphering the role of antibodies in protection from theseviruses (279). While there are no published reports of immu-nity to Ebola virus infection after a primary infection, trans-fusion of convalescent phase whole blood to infected patientsin the 1995 Kikwit outbreak was described to confer increasedresistance in treated patients (280). A panel of human MAbs


to Ebola virus Zaire was generated from immune librariesconstructed from the bone marrow lymphocytes of two donorswho recovered from infection with the Kikwit Ebola virus in1995. Several Fabs were selected from two independentimmune libraries of 6� 106 and 2.2� 106 clones from marrowof two individuals, and another with diversity of 5� 106 fromthe PBLs of 10 donors. Binding clones were selected usinggamma-irradiated whole virus or infected cells as the selec-tive antigen, and immunoprecipitation to determine reactiv-ity to either the nucleoprotein or the envelope glycoprotein.The specificity of the binding Fabs to Ebola virus was con-firmed using immunofluorescence to detect binding to liveand to fixed Ebola virus-infected cells. One of the Fabs tothe envelope glycoprotein, KZ52, neutralized Ebola virus inboth the recombinant Fab and recombinant whole humanIgG form. Of note, one of the nucleoprotein-specific Fabs wasinhibited from binding by 10 of 10 seropositive convalescentdonor serum. This suggests that this Fab may be useful asa serological diagnostic assay in a C-ELISA for Ebola inhumans and possibly other species (279). The inability ofequine immune serum, produced against whole inactivatedEbola virus, to protect infected macacques (281) reveals thepresent limitations of polyclonal antibody preparations, andfurthermore, supports continued studies on the develop-ment of therapeutic MAbs with high specific activity againstfiloviruses.

Bacterial exotoxins provide an exquisite example ofstringent selection and of the immediate protection affordedby passive antibody therapy. In the past, toxoids have beenused as both active vaccines and to generate equine immuneserum for some of the more common bacterial toxins. Today,there are no technological limitations holding back the repla-cement of equine immune serums with human monoclonalantibodies. Collectively, bacterial exotoxins, scorpions, spi-ders, bee venoms, stonefish, box jellyfish, and snake toxinscause a lot of human morbidity and mortality world wide eachyear. However, the diversity of these toxins makes it unlikelyor impractical to produce active vaccines for each, and againtherapeutic antibodies offer the best new hope for protection.


There are therapeutic anti-venoms produced for many ofthese toxins.

Clostridial neurotoxons are the most toxic substancesknown to man (282), with murine LD50 values ranging from0.1 to 1ng=kg of body weight. Human immune serum pro-duced against Botulinum toxin (where the toxin moiety isdenatured and thus not lethal to administer) neutralizes thetoxin in vitro compared to nonimmune serum, which doesnot (283). In cases of food poisoning, equine immune serumis still used today as a passive vaccine and protects humansfrom lethal intoxication. However, there are some adversereactions to the horse antiserum, and cleaner and more homo-geneous human MAb preparations will one day replace theequine source.

There are seven different known serotypes (A–G), andthere is evidence that recombinant combinatorial forms ofthe toxin may exist (284). The diverse genetic locations ofthe botulinal neurotoxins and recent genome sequence dataon Clostridium botulinum species collectively support thenotion that they are encoded within transposons or otherhighly mobile genetic elements (Dr. S Hayes, University ofSaskatchewan, personal communication). Neutralizingantibodies are thus needed to be able to neutralize all formsof this toxin either by targeting conserved sites or byproducing pools of MAbs containing type-specific neutralizingantibody.

Recently, neutralizing human scFvs were selected fromimmune repertoires of volunteers immunized with penta-valent botulinum toxoid (285). The immune library wasprepared from PBLs with a measurable protective titer inthe mouse serum neutralization bioassay. The scFvs wereselected from immune and nonimmune libraries of 7.7� 105

and 6.7� 107 clones, respectively, upon botulinum neurotoxin(BoNt)=serotype A. Of note, while binding clones were identi-fied from both libraries, neutralizing scFv was derived onlyfrom the immune library, but not from the nonimmune humanlibrary. Moreover, scFvs specific to each of the five toxins usedin the penta-valent vaccine were derived only from theimmune library. Neutralization of the toxin correlated with


affinity and competition with holotoxin for binding sites on theheavy chain of BoNt. Moreover, neutralization was synergisticamong some of the scFvs revealing that multiple epitopes weretargeted by the scFvs. The neutralizing scFvs selected fromthe immune library exhibited Kd values of 36.9 and 7.8nM,which are comparable to values reported from hybridomas(286). Nonimmune scFvs had lower affinities with Kd valuesranging from 460 to 26nM. This clearly demonstrates someof the advantages, which can be had by using immunelibraries. While it is likely that large nonimmune librariescan have binding clones to the solvent accessible areas of agiven toxin, it is less likely that these clones bind well to toxinslike BoNt, which contain a limited number of antigenicallyvariable protective epitopes. Immunization and selectiveexpansion directs the recognition of toxins to a limited numberof immunodominant epitopes by clones, which produce protec-tive antibodies. It is logical to rationalize that the immune sys-tem of vertebrates has evolved to include inherent reactivitytoward these protective domains of highly lethal toxins understringent selection pressure over evolutionary time.

The antibodies selected against different infections aresummarized in Table 11. First, it was demonstrated thatbroadly neutralizing MAbs to HIV-1 could be selected from

Table 11 Infections Human MAbs to Infections Pathogens fromImmune Libraries

Antigen Library (size) Kd, nMTherapeuticpotential Reference

HIV-1 gp120 Fab (1� 107) 10 Virus neutralization (267)RSV Fab (5� 107) nd Virus neutralization (336)Measles virus Fab (107) 10 nd (337)HCV Fab (3� 107) 151 nd (338)Rotavirus Fab (2.5� 107) nd nd (339)Ebola virus Fab (6� 106) nd Virus neutralization (279)Measles virus Fab (> 109) nd Virus neutralization (340)Botulinum toxin Fab (107) 7.5 Toxin neutralization (285)

Antigen abbreviations: HIV-1 gp120, human immunodeficiency virus type-1 glyco-protein 120; RSV, respiratory syncytial virus; HCV, hepatitis C virus.


the gp120-hyperstimulated B cell pool of humans. Second,human MAbs to highly pathogenic Pox viruses and filovirusescan be selected from immune repertoires of boosted B cellsdonors and convalescent bone marrow, respectively. Third,nanomolar affinity neutralizing scFvs can be obtained fromantibody libraries from immune human repertoires but notfrom naıve repertoires. The direct selection of human MAbfragments will continue to depend upon availability of conva-lescent or vaccine immunized donors.

III.C.2. Autoimmune

Rheumatoid arthritis (RA) is an autoimmune disease and cur-rently has the largest number of patients being treated withMAbs (287). Indeed, the study of human autoantibodyresponses is the field where only immune libraries can beused and there is no competition from synthetic and naıvelibraries. For instance, dsDNA-specific Fabs with moderateaffinities can be isolated from libraries prepared from healthyand systemic lupus erythematosus (SLE) donors. However,high-affinity Fabs were isolated only from an SLE library(268). This is even more significant for some autoantibodiesfor which the frequency of specific B cell precursors is verylow. The B cell precursor frequency for SLE-specific anti-Smith antibodies (anti-Sm) has been shown to be less than1:30,000 splenocytes in the autoimmune mouse model (288).The human anti-Sm IgG autoantibodies were successfullygenerated and characterized from an SLE patient (288,289).Taking into account the limitations that conventional hybri-doma technology impose on the generation of human MAbs,the phage display approach seems to be the main technologycapable of generating functional human autoantibodies fromimmune donors. In 1994, Hexham et al. (290) first appliedthe pComb3 system to produce three thyroid peroxidase auto-antibodies from a patient with Hashimoto’s thyroiditis, thusdemonstrating that the phage display system is an effectiveway to produce recombinant autoantibodies.

Because combinatorial antibody libraries randomyrecombine heavy and light chains, the issue whether the


selected antibodies are disease-revelant autoantibodiesshould be addressed. In 1996, Roben et al. reported thatanti-dsDNA autoantibodies could only be recovered from thelibrary of an SLE patient and not from a library from ahealthy identical twin of the patient (291). That study sug-gested that in combinatorial libraries the de novo pairing ofheavy and light chains unrelated to the in vivo autoimmuneresponse did not create the high-affinity disease-associatedautoantibodies. More recently, Jury et al. demonstrated thatnatural Ig heavy and light chain pairings of autoantibodiescould be isolated from two patients at onset of type 1 diabetes(292). An IgG1kl library of 2� 106 independent clones wasconstructed from PBLs of two diabetic patients. After fiveround of panning on glutamate decarboxylase (GAD65), oneof the major autoantigens, eight GAD65-reactive clones wereisolated. Three of them reflected all typical features of natu-rally occuring GAD65 autoantibodies. Sequence comparisonto monoclonal islet cell antibodies (MICA) demonstrated thatthe heavy chain of GAD65ab was identical and light chainwas nearly identical to corresponding chain of MICA6. Thus,the authors demonstrated for the first time that selectedclones reflect well the natural autoantibody response in type1 diabetes.

Although the generation of human monoclonal autoanti-bodies is critical for understanding humoral immune responsein autoimmunity, we would like to discuss two diseases wherethe selected autoantibodies have therapeutic potential. In thefirst example, Zeidel et al. (288) first reported the generationof functional human autoantibodies from PBLs of patientwith myastenia gravis (MG) (288). Using the pComb3 vector,an IgG1k library was constructed and panned against purifiedacetyl choline receptor (AChR). After five rounds of panning,four positive clones were selected and all demonstrated theability to stain the muscle cells. The therapeutic potential ofanti-AChR antibodies was demonstrated 2 years later byGraus et al. (293). The IgG1k and IgG1k libraries were con-structed from thymic tissue obtained immediately after ther-apeutic thymectomy. Panning was performed using humanAChR expressed by TE671 rhabdomyosarcoma cells. Four


different clones were selected after five round of panning andall four Fabs stained human AChR expressed by the TE671cells. More importantly, Fab 637 inhibited the binding ofserum anti-human AChR from MG patient by more than90%, whereas the combination of Fabs 637 and 587 was ableto limit AChR loss induced by MG serum to 20%. Since therecombinant anti-humanAChR Fabs do not interfere withreceptor function and are unable to activate complement, itis feasible that they might be used to protect the AChRagainst degradation by intact autoantibodies in vivo duringmyastenic crisis.

In 1995, Ishida et al. first successfully selected recombi-nant human Fab against integrin aIIb3 from a phage librarygenerated from PBLs of a patient with Glanzmann thrombas-tenia (GT). Integrin aIIb3 is highly immunogenic in humansand remains the most frequently identified target of humanautoantibodies that have been detected in a majority ofpatient with autoimmune thrombocytipenic purpura (AITP)and in patients with GT (294,295). Ishida et al. (296) demon-strated a restricted usage of the VH4 gene family in theselected Fabs. In this context, the knowledge of the geneticstatus of anti-aIIb3 MAbs could help advance our understand-ing of the pathogenesis of autoantibody development. Jacobinet al. (297) adapted antibody library technology to determinethe nature of the humoral immune response in patients withAITP and GT. Two scFv IgG1kl libraries were constructedfrom PBLs of GT patients and from spleen tissue of AITPpatients. Several positive scFv clones were selected aftertwo rounds of selection on activated platelets. After confirm-ing the specificity of the selected clones on aIIb3 by ELISA,the scFv-binding affinities of two clones, TEG4 and EBB3,were determined by C-ELISA. The TEG4 exhibited a Kd valueof 2.6� 10�6M and EBB3 exhibited a Kd value of 1.8�10�7M. The nucleotide sequence of variable regions of theselected clones revealed a polyclonal response in both patients.A large repertoire of VH and VL genes was used. The selectedfully human Fabs reported in this article might be more suita-ble for repeated therapy as antagonist of aIIb3 integrin thanthe most commonly used chimeric Fab2’ 7E3.


The antibodies selected from autoimmune libraries aresummarized in Table 12. We have summarized this sectionwith the following points. First, it was demonstrated thathigh-affinity dsDNA-specific antibodies could be isolated forpatients with SLE. Second, anti-thyroid peroxidase MAbscan be recovered from immune libraries made from the B cellsof patients with Hashimoto’s thyroiditis. Third, naturallypaired autoantibodies to GAD65 can be isolated from patientswith early onset of type 1 diabetes. Fourth, anti-AChR MAbscan be selected from immune libraries of patients with MG.Fifth, scFv MAb fragments with low nanomolar affinitiescan be selected from immune libraries of AITP patients. Theability to generate relevant human autoimmune MAbs byphage display allows investigators to define the antigenic epi-topes targeted by autoimmune responses as well as to under-stand the genetic and structural bases of pathogenicautoantibody responses. Phage display has been used success-fully to produce human autoantibodies from patients withdiverse spectrums of autoimmune diseases, such as Hashimo-to’s thyroiditis (253), SLE (291), ulcerative colitis (298), idio-patic dilated cardiomyopathy (299), and autoimmunegastritis (300). Some of the selected autoantibodies were inagreement with previous studies that suggested that theVH4 family is a major source of autoantibodies (301). Theseand many other publications have greatly advanced our

Table 12 Autoimmune

Antigen Library (size) Kd, nM Therapeutic potential Reference

TPO Fab (105) 1 nd (290)dsDNA Fab (8� 106) 7.6 nd (268)Tg Fab (5.3� 107) 2.8 nd (253)AChR Fab (1.1� 106) nd Receptor protection (293)Sm Fab (2� 107) 10 nd (289)GAD65 Fab (2� 106) nd nd (292)aIIB3 scFv (1.5� 107) 180 Integrin antagonist (297)

Antigen abbreviations: aIIB3, integrin aIIB3; dsDNA, double strand DNA; Sm, anti-Smith; TPO, thyroid peroxidase; GAD65, decarboxylase; AChR, acetyl choline recep-tor; Tg, thyroglobuline.


understanding of the pathogenesis of antibody-mediatedautoimmune diseases.

III.C.3. Tumors

There are today five MAbs approved by the FDA for the treat-ment of cancer (Table 10), and numerous others are inadvanced clinical development (302,303). It is thought thatthe humoral antibody response in cancer patients may beselectively directed toward the ‘‘nonself ’’ antigens expressedby the autologous tumor cells. Indeed, a humoral immuneresponse directed to known tumor antigens has been demon-strated by the presence of serum antibodies in patients withcancer (252). Thus, B cells from sensitized cancer patientswho have mounted a response to altered or over expressedantigens are a valuable source of mRNA to construct phage-displayed antibody libraries (67). These cancer patients mayprovide an enriched source of disease-related antibodies thatcan be recovered from antibody libraries sorted by panningagainst specific tumor antigens. For example, a humoralimmune response to tumor-related antigens such as p53 anderbB-2 has been demonstrated in some patients with cancer(304,305). Additionally, it is of fundamental importanceto develop new immunotherapeutic protocols designed todetermine which antigens are the target of an immuneresponse in the various types of cancer. This will lead tonew therapies and treatments in particular for those cancerswith poor prognoses.

Tumor-specific human-MAb fragments have been iso-lated against melanoma. In the first report, tumor-specificMAb fragments were isolated from individuals immunizedwith interferon-gamma (IFNg)-transduced autologous mela-noma cells (306). The PBLs were isolated from two melanomapatients immunized with in vitro cultured, autologous-tumorcells infected with a retroviral vector carrying the humanINFg gene. Three scFv libraries were constructed using thefUSE5 phage vector for multivalent scFv display; the smallestof these two libraries contained 4� 107 clones. These scFvfragments were synthesized from both the IgM and IgG,and l and k light chain classes of mRNA. The selection


involved panning against the autologous melanoma cell line,followed by extensive absorption against melanocites toincrease the chance of isolating antibodies with specificityfor tumor. After three panning cycles, the majority of theselected clones reacted with all melanoma cell lines in ELISAand therefore recognize common melanoma antigens. How-ever, clone V86 showed the tightest association with mela-noma lines and demonstrated an intense staining ofmelanoma tissue but did not react with normal tissues pre-sent in the melanoma sections. This MAb clone is a poten-tially important tool for diagnostic screening for melanoma.The selection technique itself can be used to screen the anti-body repertoire of any cancer patient, which may provideaccess to many more human antitumor MAbs.

Others have used a similar positive=negative selectionstrategy for antibody phage against melanoma. The selectiveremoval of clones that react with normal tissue obviates theneed for pure antigen (307). For example, a human scFv IgGklibrary (9.3� 107 clones) was constructed in pCANTAB5 vec-tor from PBLs of 10 donors with a high titer of autoantibodies.After three rounds of using positive=negative selection strat-egy, they isolated two scFv clones, B3 and B4, that were posi-tive on all melanoma sections obtained from several differentpatients. Neither B3 nor B4 crossreacted with normal tissues,except for the weak reactivity of B3 with normal liver. ThescFv clones reacted with antigens that are also expressed ontumors other than melanoma. Therefore, this approachshould be applicable to the isolation of human antibodiesagainst tumor markers or novel cell surface markers in gen-eral. The human antitumor scFvs B3 and B4 have a diagnos-tic value for the detection of metastatic disease byradioimaging of patients. Another potential application wouldbe the use of these recombinant antibodies in targeted drugdelivery to specifically kill the tumor cells in vivo.

Human MAb fragments against c-erbB-2 have beenisolated from immune libraries of patients with colorectalcancer. Clark et al. (308) described the isolation of Fabs toc-erbB-2 from an immune library constructed from theisolated pericolic lymph node of a colorectal cancer patient.


An IgG1k library contained 2� 107 clones and was biopannedwith purified recombinant c-erbB-2. After three rounds ofselection, 16 clones showed a unique restriction pattern andreproducible reactivity against c-erbB-2 protein. Five Fabswith good expression levels were futher assessed for immu-noreactivity against tumor cell lines. Remarkably, theseMAbs bound strongly to the c-erbB-2 positive cell line,SKBR3, but displayed no significant staining of the negativeMDA-MB-231 cells. All of the anti-c-erbB-2 Fabs isolated inthis study used commonly represented VH genes (VH4, VH3,and VH1) with apparent VH4 over-representation, which sug-gests clonal selection. These results indicate that a naturallyoccuring immune response to tumor-related antigens can beexploited with phage display libraries for understanding ofimmune response to tumor cells and for the isolation of Fabsto predefined target antigens.

Human MAb fragments against p53 have also beenisolated from immune libraries of patients with colorectalcancer. The same group selected anti-p53 Fabs from IgG1klibraries also contructed from a similar library made fromthe pericolic lymph nodes taken from six colorectal cancerpatients (252). After five rounds of panning against recombi-nant p53, 14 unique clones were isolated from one library of4.5� 107 clones. The selected Fabs were encoded by germ-line (unmutated) V genes, predominantly from the VH1

family. Four of these Fabs were purified and further ana-lyzed in detail. All four recognized p53 in ELISA andshowed no reactivity against other antigens includingErbB2, MUC-1, CEA, TT, insulin, KLH, and BSA. InhibitionELISA using serum from donor patients demonstrated var-ious degrees of inhibition for each of the Fabs. The Fab163.1 had high affinity for p53 and exhibited a Kd of11.9nM by surface plasmon resonance analysis. As thereare no other human anti-p53 MAbs available from conven-tional cell immortalization, the information gained onhuman anti-p53 antibody V gene usage using immune anti-body libraries, specifically regarding the degree of somaticmutation, is critical to any understanding of the natureand significance of the humoral immune response to p53.


The Fabs selected in this study also have advantage overmurine MAbs when used as anti-idiotypic vaccines. In2001, the same group isolated a humanFab against the centralDNA-binding domain of p53 from a large IgG1-4kl immuneantibody library (309). This Fab, 1159.8, provides additionalinsight into thenature of the tumor-specific immune responsesand can be used as a reagent for functional studies of p53. Ofnote, this study also revealed that Fab 1159.8 binds to an epi-tope in close proximity to murine MAb Fab 240 that was ableto prevent tumor growth and metastasis upon immunization(310). Therefore, Fab 1159.8 may prove useful as an antitu-mor idiotypic vaccine while avoiding the undesirable featuresof the murine MAb.

Other studies have been performed using antibodylibraries from colorectal cancer patients to investigate thenature and specificity of the humoral immune response. Thelymphocytes infiltrating the primary colorectal tumor andlymph nodes draining the tumor were used for the construc-tion of a Fab IgG1kl antibody libraries containing greater that108 clones (311). The antibody repertoires constructed fromthese two tissues were screened for the presence of antibodiesdirected to colorectal cancer cells by cell-based selection uponthe cancer cell line CaCo2. For comparison, the same selec-tions were performed with a phage antibody repertoire madefrom B cells of healthy donors, which would in this case repre-sent the relative ‘‘naıve’’ library (76). Striking differenceswere observed in the panel of specificities selected from thesedifferent repertoires. Although a large panel of antibodiesreactive with patient-derived primary tumors was obtainedfrom the immune repertoires after two round of panning,antibodies selected from the local immune sources were direc-ted to intracellular (cytoplasmic and nuclear) targets only.However, selections using the nonimmune library did resultin numerous antibodies that recognized cell surface markerson CaCo2. Although these data do not rule out the existenceof humoral responses to certain cell surface antigens, theysuggest a bias in the local humoral immune response in thiscolorectal cancer patient, directed primarily toward intracel-lular target antigens.


High-affinity human MAbs against the LewisX antigenhave been isolated from immune repertoires. In 1999, an anti-body library derived from the PBLs of 20 patients with var-ious cancer diseases was successfully exploited to isolatescFv antibodies specific for the carbohydrate antigens sialylLewisX (sLeX) and LewisX (LeX) (312). An scFv antibodylibrary was prepared from the assembly of VH, Vk, and Vl

PCR-amplified gene pools to yield approximately 2� 108

clones. After four round of panning, four unique scFvs werethen selected by using synthetic sLeX and LeX BSA conju-gates. The selected scFv fragments were specific for sLeX

and LeX, as demonstrated by ELISA, BIAcore, and flow cyto-metry binding to the cell surface of pancreatic adenocarci-noma cells. The Kd value of the best sLeX binder, S6, wasequal to 110nM, which is comparable to the affinities of MAbsnormally derived from the secondary immune response.These selected scFvs could be valuable reagents for probingthe structure and function of carbohydrate antigens and inthe treatment of human tumor diseases.

High-affinity human MAbs against the ganglioside GM3

antigen have been isolated from immune repertoires. Thesame multi-etiology library was used again to select scFvsagainst ganglioside GM3 overexpression, which is associatedwith a number of different cancers, including skin, colon,breast, and lung (313). Several scFvs were affinity selected.One scFv, GM3A8, was purified and found to exhibit a Kd of1200nM by surface plasmon resonance analysis. The in vitroaffinity of the scFv was estimated to be comparable to a pre-viously reported GM3-binding murine IgG and anti-GM3 poly-conal antibodies (314,315), but with the significantadvantages of possessing a human antibody sequence andhaving the ability to specifically recognize highly metastaliccancer cells vs. normal cells. The GM3A8, with its favorablebinding properties for melanoma and breast tumor cells, pro-vides a solid foundation for further development.

The antibodies selected from cancer patient libraries aresummarized in Table 13. We have summarized this sectionwith the following points. First, it was demonstrated thatautologous cells can be used in antibody selection strategies,


thus circumventing a major weakness of the phage displaytechnique such as the requirement of pure antigen for phageselection (306,307). Second, anti-c-erbB-2 Fabs can be sele-cted from immune libraries made from B cells of colorectalcancer patients (308). Third, nanomolar affinity antip53 Fabscan be selected from immune libraries made from B cells ofcolorectal cancer patients (252). Fourth, a large bias towardintracellular antigens was demonstrated in the local humoralimmune response in a colorectal cancer patient (311). Finally,high-affinity human antibodies against several tumor-asso-ciated carbohydrate antigens could be selected from a phagelibrary constructed from the PBLs of various cancer patients(312,313). That approach did not depend on the necessity torepeatedly construct phage antibody libraries.

III.D. Nonhuman Primates

The primate immune system and B cell repertoire are highlysimilar to that of humans. Therefore, everything discussedin the human section (see above) about the immune repertoireand primer design applies for primates as well. The closegenetic relationship between nonhuman primates andhumans make primates the paramount species for develop-ment of therapeutic antibodies for use in humans. However,research into this area has been severely limited due to the

Table 13 Tumors Human MAbs to Tumor Targets from ImmuneLibraries


Melanoma scFv (2� 108) nd Diagnostic drug/delivery (306)ErbB2 Fab (2� 107) nd Immunotherapy (308)Melanoma scFv (9.3� 107) nd Diagnostic drug/delivery (307)sLeX scFv (2� 108) 110 Drug delivery (312)p53 Fab (4.5� 107) 11.9 Anti-Id vaccine (252)p53 Fab (1.6� 107) nd Anti-Id vaccine (309)CaCo2 Fab (> 108) nd nd (311)GM3 scFv (2� 108) 1200 Drug delivery (313)


enormous expense associated with breeding colonies and theprice of a single primate, for example a chimpanzee is quitehigh. Moreover, there remains a lot of public scrutiny as wellas criticism from private groups, which is invariably asso-ciated with research involving primates. Several publicationshave tried to explore the substitution of primate serologicalproducts in human prophylaxis and therapy by introducinghuman immune molecules into primates (316–318). Collec-tively, these data demonstrate that human immune moleculesare essentially nonimmunogenic when introduced to chimpan-zees relative to other primates, and that human antibodies arerecognized as being ‘‘self ’’ by the chimpanzee immune system.Indeed, the half-life of a human MAb in a chimpanzee wasfound to be equivalent to the estimated half-life of IgG inhumans (318). Conversely, it is likely that chimpanzee antibo-dies would be like ‘‘self ’’ in humans or at least nonimmuno-genic and useful without further modifications (for exampleto glycosylation patterns) in human therapy (81).

In the first example, Glamann et al. (319) produced anantibody phage display library constructed from RNAextracted from lymph node cells of a SIV-infected long-termnon-progressor macaque (rhesus). The Fab library was fully‘‘macaque-like’’ in that even the Ck and the CH1 domains werecloned from macaques using human primers. From thislibrary with a primary diversity of 3� 107 clones, sevengp120-reactive Fabs were obtained by selection of the libraryagainst SIV monomeric gp120. Although each of the Fabs wasunique in molecular sequence, they all had highest homologyto the human VH4 gene family. Furthermore, they formed twodistinct groups based on epitope recognition, neutralizingactivity in vitro, and molecular analysis. The first group ofFabs did not neutralize SIV and bound to a linear epitope inthe V3 loop of the SIV envelope. In contrast, two of the group2 Fabs neutralized homologous, neutralization-sensitiveSIVsm isolates with high efficiency but failed to neutralizeheterologous SIVmac isolates. Based on C-ELISAs withmouse MAbs of known specificity, these Fabs reacted witha conformational epitope that includes domains V3 and V4of the SIV envelope. These macaque Fabs not only provide


valuable standardized and renewable reagents for studyingthe role of antibody in SIV-associated disease, but they alsoset the stage for future use of macaques to develop MAbsfor use in humans therapy and prophylaxis.

In the second example, Schofield et al. (81) produced aFab phage display library constructed from the RNA extractedfrom bone marrow lymphocytes of a chimpanzee, which hadbeen previously experimentally infected with all five hepati-tis-causing viruses, hepatitis A, B, C, D, and E. The Fablibrary IgG1k was made with a primary diversity of 1.9� 107

clones using again human Ck and CH1 domain primers. Twohepatitis E virus (HEV) ORF2 capsid protein-reactive Fabswere obtained by selection of the library against SAR-55ORF2. Competition experiments revealed that both Fabsreacted to the same epitope with affinities in the single-digitnanomolar range. These Fabs had highest homology to thehuman VH3 family and both neutralized the SAR-55 strain ofHEV in vitro as shown by the complete prevention of infectionof chimps inoculated with live HEV following preincubationwith either Fab. Despite the high cost of using chimpanzeesas a donor for immune repertoires, there are several advan-tages (81). The first is that chimpanzees can be infected withmany important human viral pathogens and the chimp isthe most closely related primate to humans, and thus, chim-panzee antibodies may be able to be used directly in immuneprophylactic treatment of infectious disease.

Antibody libraries from immune repertoires can be usedto select MAbs against multiple targets if the lymphocyteswere sensitized to these targets prior to tissue collection.Alternatively, the library can be used to select for binders toantigens conserved between strains of variant organisms.The same antibody library, which was used in the hepatitisE example (above), was used to select MAbs against hepatitisA virus (HAV). Two years later, the authors reported usingwhole inactivated HAV particles as the panning antigen toisolate four Fabs to the HAV capsid (320). Following threerounds of panning on HAV particles, four unique HAV-specificclones were identified. Isolated Fabs competed with one of themurine MAbs that were used to define the HAV antigen site.


These results seem to be in accordance with the earlier studiesthat suggested that there is a single immunodominant anti-genic site on the HAV capsid. Overall, the Fab library usedto isolate MAbs to both HEV and HAV is a potential repositoryfor antibodies to all five recognized human hepatitis viruses,since the donor chimpanzee had been infected with each ofthe hepatitis viruses. Such chimpanzee-derived Ig sequencesdiffer from human-derived sequences no more than geneti-cally distinct human sequences differ from each other, andthus they have direct therapeutic potential.

Melanoma-specific scFvs have been derived from nonhu-man primates. Two Cynomolgus monkeys were immunizedwith a crude suspension of metastatic melanoma (321). AnscFv antibody phage library was generated from the lymphnode mRNA with approximately 3� 107 primary clones.Several clones producing scFvs that reacted with melanomaantigens were identified after three rounds of panning usingmelanoma cells and tissue sections. One of these scFvs,K305, demonstrated high-affinity binding and selectivity,supporting its use for tumor therapy in conjunction with Tcell-activating superantigens. Comparison with the humangermline sequence of the Vl and VH genes demonstrated89% and 92% sequence identity on the nucleic acid level.Clone K305 was fused as a primate Fab to staphylococcalenterotoxin A. The affinity for melanoma tissue was in thelow to subnanomolar range. T cell-mediated lysis of mela-noma cells and in vivo tumor reduction mediated by thisantibody in SCID mice were demonstrated, suggesting applic-ability for immunotherapy of malignant melanoma.

The antibodies selected from nonhuman primatelibraries are summarized in Table 14. We conclude this sec-tion with the following points. First, it was demonstrated thatspecific MAbs could be selected against enveloped virusesfrom immunized macaques. Second, MAbs from chimpanzeeswere selected against multiple hepatitis virus strains bysimply including these in the initial immunization strategy.Third, cancer cell-specific MAb fragments can be derived fromimmune antibody libraries from nonhuman primates. Thesesuccessful examples of the use of antibody libraries from


nonhuman primates for the development of potential immu-notherapeutic interventions for infectious diseases andtumors are rare, but they are expected to continue into thefuture as primate antibodies are extremely similar tohuman antibodies and can rapidly translate into clinicaltreatments.

IV. THE FUTURE

Antibody libraries from immune repertoires have developedinto a new and exciting fieldwith broad applications in biotech-nology and the pharmaceutical industry.Use of this technologyin centralized laboratories will be crucial to the development ofmodern and rapid diagnostics for confirmatory diagnostictests. These tests will, in the next decade, be transformed intosmall high-throughput antigen detection chips, where a singletest can screen for thousands of pathogens.However, the key todeveloping these protein-chips and other nanodetection tech-nology is to first develop quality-controlled reagents capableof detecting pathogenic proteins. Clearly antibody librariesfrom immune sources will be key to the production of manyof the toxin, pathogen, autoantigen, and tumor-specific antibo-dies required for this technology. B cell immortalization techni-ques and phage display offer complementary approaches to thedevelopment of antigen-specific MAbs. It has become increas-ingly important for scientists in many fields including neuro-logy, cancer biology, autoimmunity, placentology, infectious

Table 14 Nonhuman Primates MAbs from Immune Libraries


SIV gp120 Fab (3� 107) nd Virus neutralization (319)HEV ORF2 Fab (1.9� 107) 1.7 Virus neutralization (81)Melanoma scFv (3� 107) 1.6 Drug delivery (321)HAV Fab (1.9� 107) nd Virus neutralization (320)

Antigen abbreviations: SIV gp120, simian immunodeficiency virus glycoprotein 120;HEV ORF2, hepatitis E virus open reading frame 2 protein; HAV, hepatitis A virus.


disease, and pure biotechnology to be capable of alternatingbetween hybridoma growth in tissue culture and antibodycloning in order to completely capitalize on new leadmoleculesfrom immunized sources.

The development of MAbs by traditional methods con-tinues to evolve. New high-throughput screening systemsare on the horizon for hybridoma and E. coli produced MAbs.The generation of antibody libraries from immune repertoiresprovides a unique tool for fundamental research on the immu-nology of human and animal diseases. Moreover, immunelibraries and the power of phage selection allow for the selec-tion of antibodies to complex antigens.

Antibody libraries will continue to be developed anewfrommore diverse species such as fish, as genomic informationabout Ig repertoires is unveiled. For example, a naturallyoccurring VH-like domain, with characteristics similar tocamel VHH antibodies (322), has recently been described innurse sharks as the new antigen receptor (NAR) (323). TheNARs from wobbegong shark have recently been used asscaffolds for the construction of phage-displayed libraries(324). Similarly, new inroads have been made in the geneticsof Atlantic salmon.

The recent cloning and sequencing the cDNA of around50 VH (VDJ) and 15 VL genes have quickly led to the selec-tion of anti-TNP and anti-FITC-specific scFvs (325). Thiswork opens the possibility of using the immune repertoiresof fish for MAb generation.

We predict a rapid growth in the area of infectious dis-eases given the huge problem related to antibiotic resistanceand the need for new therapies. There is actually a remark-able death of human MAbs in clinical trials to infectiousagents. Indeed, only 7 of 128 MAbs in clinical trial areagainst infectious agents (77). The vast majority of theseclinical antibodies are against cancer targets. There are alsovery few MAbs from animals for validating diagnoses of ani-mal diseases. This seems counter-logical given the fact thatantibodies evolved in order to target foreign antigens.However, this is more likely a reflection of first world healthconcerns.


The use of immune libraries from xenomice will be idealsource of cDNA for in vitro modifications including affinitymaturation or humanization of existing potent murine MAbs,once again highlighting the complementary nature of thesetwo techniques. The exquisite ability of the immune systemto produce antibodies to a specific immunogen in vivo,whether on a pathogen, a cryptic self antigen, or an overex-pressed or altered tumor antigen, will ensure that antibodylibraries from immune repertoires will continue to beexploited in the future.

ACKNOWLEDGMENTS

The authors would like to thank Dev sidhu for his time andpatience. JDB is supported by CBRN research and technologyinitiative.

REFERENCES

1. Groves DJ, Morris BA. Veterinary sources of nonrodent mono-clonal antibodies: interspecific and intraspecific hybridomas.Hybridoma 2000; 19:201–214.

2. Flynn JN, Harkiss GD, Hopkins J. Generation of a sheep �mouse heterohybridoma cell line (1C6.3a6T.1D7) andevaluation of its use in the production of ovine monoclonal anti-bodies. J Immunol Methods 1989; 121:237–246.

3. Hoogenboom HR, Chames P. Natural and designer binding sitesmade by phage display technology. Immunol Today 2000;21:371–378.

4. Rader C. Antibody libraries in drug and target discovery. DrugDiscov Today 2001; 6:36–43.

5. Soderlind E, Strandberg L, Jirholt P, Kobayashi N, Alexeiva V,Aberg AM, Nilsson A, Jansson B, Ohlin M, Wingren C, Daniels-son L, Carlsson R, Borrebaeck CA. Recombining germline-derived CDR sequences for creating diverse single-frameworkantibody libraries. Nat Biotechnol 2000; 18:852–856.

6. Knappick A, Ge L, Honegger A, Pack P, Fischer M, WellnhoferG, Hoess A, Wolle J, Pluckthun A, Virnekas B. Fully synthetic


human combinatorial antibody libraries (HuCAL) based onmodular consensus frameworks and CDRs randomized withtrinucoleotides. J Mol Biol 2000; 296:57–86.

7. Smith G. Filamentous fusion phage: novel expression vectorsthat display cloned antigens on the virion surface. Science1985; 228:1315–1317.

8. Parmley SF, Smith GP. Antibody-selectable filamentous fdphage vectors: affinity purification of target genes. Gene 1988;73:305–318.

9. Scott JK, Smith GP. Searching for peptide ligands with an epi-tope library. Science 1990; 249:386–390.



12. Skerra A, Pluckthun A. Assembly of a functional immunoglo-bulin Fv fragment in Escherichia coli. Science 1988; 240:1038–1041.

13. Better M, Chang CP, Robinson RR, Horwitz AH. Escherichiacoli secretion of an active chimeric antibody fragment. Science1988; 240:1041–1043.

14. Chiang YL, Sheng-Dong R, Brow MA, Larrick JW. DirectcDNA cloning of the rearranged immunoglobulin variableregion. Biotechniques 1989; 7:360–366.

15. Gavilondo-Cowley JV, Coloma M, Vazquez J, Ayala M, MaciasA, Fry K, Larrick J. Specific amplification of rearranged immu-noglobulin variable region genes from mouse hybridoma cells.Hybridoma 1990; 9:407–417.

16. Coloma MJ, Larrick JW, Ayala M, Gavilondo-Cowley JV.Primer design for the cloning of immunoglobulin heavy-chainleader-variable regions from mouse hybridoma cells usingthe PCR. Biotechniques 1991; 11:152–156.

17. Huse WD, Sastry L, Iverson S, Kang A, Alting-Mees M, BurtonDR, Benkovic SJ, Lerner R. Generation of a large combinatorial


library of the immunoglobulin repertoire in phage lambda.Science 1989; 246:1275–1281.

18. Caton A, Koprowski H. Influenza virus hemagglutinin specificantibodies isolated from a combinatorial expression library areclosely related to the immune response of the donor. Proc NatlAcad Sci USA 1990; 87:6450–6454.

19. Persson MA, Caothien RH, Burton DR. Generation of diversehigh-affinity human monoclonal antibodies by repertoire clon-ing. Proc Natl Acad Sci USA 1991; 88:2432–2436.


21. Clackson T, Hoogenboom HR, Griffiths AD, Winter G. Makingantibody fragments using phage display libraries. Nature1991; 352:624–628.

22. Kang AS, Barbas CF, Janda KD, Benkovic SJ, Lerner RA.Linkage of recognition and replication functions by assemblingcombinatorial antibody Fab libraries along phage surfaces.Proc Natl Acad Sci USA 1991; 88:4363–4366.

23. Hoogenboom H, Griffiths A, Johnson K, Chiswell D, Hudson P,Winter G. Multisubunit proteins on the surface of filamentousphage: methodologies for displaying antibody Fab heavy andlight chains. Nucleic Acids Res 1991; 19:4133–4137.

24. Barbas CF III, Kang A, Lerner RA, Benkovic S. Assembly ofcombinatorial antibody libraries on phage surfaces: the geneIII site. Proc Natl Acad Sci USA 1991; 88:7978–7982.

25. Marks JD, Hoogenboom HR, Bonnert TP, McCafferty J, Grif-fiths AD, Winter G. By-passing immunization. Human antibo-dies from V gene libraries displayed on phage. J Mol Biol 1991;222:581–597.

26. Barbas CF III, Crowe JE Jr, Cababa D, Jones TM, Zebedee SL,Murphy BR, Chanock RM, Burton DR. Human monoclonalFab fragments derived from a combinatorial library bind torespiratory syncytial virus F glycoprotein and neutralize infec-tivity. Proc Natl Acad Sci USA 1992; 89:10164–10168.

27. Burton D, Barbas CF III. Human antibodies from combina-tional libraries. Adv Immunol 1994; 57:191–280.


28. Winter G, Griffiths A, Hawkins R, Hoogenboom H. Makingantibodies by phage display technology. Annu Rev Immunol1994; 455:12–433.

29. Barbas CF, Burton D, Scott J, Silverman G, eds. PhageDisplay: A Laboratory Manual. Cold Spring Harbour, NewYork: Cold Spring Harbour Laboratory Press, 2001.

30. Boder ET, Wittrup KD. Yeast surface display for screeningcombinatorial polypeptide libraries. Nat Biotechnol 1997; 15:553–557.

31. Schaffitzel C, Hanes J, Jermutus L, Pluckthun A. Ribosomedisplay: an in vitro method for selection and evolution of anti-bodies from libraries. J Immunol Methods 1999; 231:119–135.

32. Daugherty PS, Olsen MJ, Iverson BL, Georgiou G. Develop-ment of an optimized expression system for the screening ofantibody libraries displayed on the Escherichia coli surface.Protein Eng 1999; 12:613–621.

33. Azzazy M, Highsmith W Jr. Phage display technology: clinicalapplications and recent innovations. Clin Biochem 2002;35:425–445.

34. Golub R, Fellah JS, Charlemagne J. Structure and diversity ofthe heavy chain VDJ junctions in the developing Mexican axo-lotl. Immunogenetics 1997; 46:402–409.

35. Dammers PM, Kroese FG. Evolutionary relationship betweenrat and mouse immunoglobulin IGHV5 subgroup genes(PC7183) and human IGHV3 subgroup genes. Immunoge-netics 2001; 53:511–517.

36. Wagner B, Greiser-Wilke I, Wege AK, Radbruch A, Leibold W.Evolution of the six horse IGHG genes and correspondingimmunoglobulin gamma heavy chains. Immunogenetics2002; 54:353–364.

37. Momoi Y, Nagase M, Okamoto Y, Okuda M, Sasaki N, WatariT, Goitsuka R, Tsujimoto H, Hasegawa A. Rearrangements ofimmunoglobulin and T-cell receptor genes in canine lympho-ma=leukemia cells. J Vet Med Sci 1993; 55:775–780.

38. Patel M, Selinger D, Mark GE, Hickey G, Hollis G. Sequence ofthe dog immunoglobulin alpha and epsilon constant regiongenes. Immunogenetics 1995; 41:282–286.


39. Cho KW, Satoh H, Youn HY, Watari T, Tsujimoto H, O’BrienSJ, Hasegawa A. Assignment of the cat immunoglobulin heavychain genes IGHM and IGHG to chromosome B3q26 and T cellreceptor chain gene TCRG to A2q12?q13 by fluorescence insitu hybridization. Cytogenet Cell Genet 1997; 79:118–120.

40. Najakshin AM, Belousov ES, Alabyev BY, Christensen J,Storgaard T, Aasted B, Taranin AV. Structure of mink immu-noglobulin gamma chain cDNA. Dev Comp Immunol 1996;20:231–240.

41. Sinkora M, Sinkorova J, Butler JE. B cell development andVDJ rearrangement in the fetal pig. Vet Immunol Immuno-pathol 2002; 87:341–346.

42. Duchosal MA, Eming SA, Fischer P, Leturcq D, Barbas CF III,McConahey PJ, Caothien RH, Thornton GB, Dixon FJ, BurtonDR. Immunization of hu-PBL-SCID mice and the rescue ofhuman monoclonal Fab fragments through combinatoriallibraries. Nature 1992; 355:258–262.

43. Nguyen H, Hay J, Mazzulli T, Gallinger S, Sandhu J, Teng Y,Hozumi N. Efficient generation of respiratory syncytial virus(RSV)-neutralizing human MoAbs via human peripheral bloodlymphocyte (hu-PBL)-SCID mice and scFv phage displaylibraries. Clin Exp Immunol 2000; 122:85–93.

44. Rajewsky K. Clonal selection and learning in the antibody sys-tem. Nature 1996; 381:751–758.

45. Klinman NR, Linton PJ. The clonotype repertoire of B cell sub-populations. Adv Immunol 1988; 42:1–93.

46. Duggan JM, Coates DM, Ulaeto DO. Isolation of single-chainantibody fragments against Venezuelan equine encephalomye-litis virus from two different immune sources. Viral Immunol2001; 14:263–273.

47. Gherardi E, Milstein C. Original and artificial antibodies. Nat-ure 1992; 357:201.

48. Kettleborough CA, Ansell KH, Allen RW, Rosell-Vives E, Gus-sow DH, Bendig MM. Isolation of tumor cell-specific single-chain Fv from immunized mice using phage–antibody librariesand the re-construction of whole antibodies from these anti-body fragments. Eur J Immunol 1994; 24:952–958.


49. Ames RS, Tornetta M, McMillan L, Kaiser K, Holmes S,Applebaum E, Cusimano D, Theissen T, Gross M, Jones C, Sil-verman C, Porter T, Cook R, Bennet D, Chaiken IM. Neutra-lizing murine monoclonal antibodies to human IL-5 isolatedfrom hybridomas and a filamentous phage Fab display library.J Immunol 1995; 154:6355–6364.

50. Krebber A, Bornhauser S, Burmester J, Honegger A, WilludaJ, Bosshard HR, Pluckthun A. Reliable cloning of functionalantibody variable domains from hybridomas spleen cell reper-toires employing a reengineered phage display system. JImmunol Methods 1997; 201(1):35–55.

51. Larrick J, Danielsson L, Brenner C, Abrahamson M, Fry K,Borrebaeck C. Rapid cloning of rearranged immunoglobulingenes from human hybridoma cells using mixed primers andthe polymerase chain reaction. Biochem Biophys Res Commun1989; 160:1250–1256.

52. Orlandi R, Gussow DH, Jones PT, Winter G. Cloning immuno-globulin variable domains for expression by the polymerasechain reaction. Biotechnology 1989; 1992:527–531.

53. Dattamajumdar AK, Jacobson DP, Hood LE, Osman GE.Rapid cloning of any rearranged mouse immunoglobulin vari-able genes. Immunogenetics 1996; 43(3):141–151.

54. Gilliland LK, Norris N, Marquardt H, Tsu T, Hayden M,Neubauer M, Yelton D, Mittler S, Ledmetter J. Rapid and reli-able cloning of antibody variable regions and generation ofrecombinant single chain antibody fragments. Tissue Antigens1996; 47:1–20.

55. Orum H, Andersen PS, Oster A, Johansen LK, Riise E,Bjornvad M, Svendsen I, Engberg J. Efficient method for con-structing comprehensive murine Fab antibody libraries dis-played on phage. Nucleic Acids Res 1993; 21:4491–4498.

56. Andris-Widhopf J, Rader C, Steinberger P, Fuller R, BarbasCF III. Methods for the generation of chicken monoclonal anti-body fragments by phage display. J Immunol Methods 2000;242:159–181.

57. Janda KD, Lo L-C, Lo C-HL, Sim M-M, Wang R, Wong C-H,Lerner RA. Chemical selection for catalysis in combinatorialantibody libraries. Science 1997; 275:945–948.


58. Moreno de Alboran I, Martinez-alonso C, Barbas CF III,Burton DR, Ditzel HJ. Human monoclonal Fab fragments spe-cific for viral antigens from combinational IgA libraries.Immunotechnology 1995; 1:21–28.

59. Steinberger P, Kraft D, Valenta R. Construction of a combina-torial IgE library from an allergic patient. Isolation and char-acterization of human IgE Fabs with specificity for the majortimothy grass pollen allergen, Phl p 5. J Biol Chem 1996;271:10967–10972.

60. Boel E, Verlaan S, Poppelier MJ, Westerdaal NA, Van StrijpJA, Logtenberg T. Functional human monoclonal antibodiesof all isotypes constructed from phage display library-derivedsingle-chain Fv antibody fragments. J Immunol Methods2000; 239:153–166.

61. de Wildt R, Mundy C, Gorick B, Tomlinson I. Antibody arraysfor high-throughput screening of antibody antigen interac-tions. Nat Biotechnol 2000; 18:989–994.

62. Rader C, Cheresh DA, Barbas CF III. A phage display approachfor rapid antibody humanization: designed combinatorial Vgene libraries. Proc Natl Acad Sci USA 1998; 95:8910–8915.

63. Hudson PJ, Souriau C. Engineered antibodies. Nat Med 2003;9:129–134.

64. Adair F. Monoclonal antibodies—magic bullet or a shot in thedark? Drug Discov World 2002; 3:53–59.

65. DeBoerM, TenVoordeG, Ossendorp F, VanDuijnG, Tager JM.Requirements for the generation of memory B cells in vivo andtheir subsequent activation in vitro for the production of antigenspecific hybridomas. J Immunol Methods 1988; 113:143–149.

66. Li Y, Kilpatrick J, Whitlam GC. Sheep monoclonal antibodyfragments generated using a phage display system. J ImmunolMethods 2000; 236:133–146.

67. Williamson RA, Burioni R, Sanna PP, et al. Human monoclo-nal antibodies against a plethora of viral pathogens from sin-gle combinatorial libraries. Proc Natl Acad Sci USA 1993;90:4141–4145.

68. Berry JD, Licea A, Popkov M, Cortez X, Fuller R, Elia M, Ker-win L, Kubitz D, Barbas CF 3rd. Rapid monoclonal antibody


generation via dendritic cell targeting in vivo. Hybrid Hybrido-mics 2003; 22:23–31.

69. Wang H, Griffiths MN, Burton DR, Ghazal P. Rapid antibodyresponses by low-dose, single-step, dendritic cell-targetedimmunization. Proc Natl Acad Sci USA 2000; 97:847–852.

70. Roost H, Bachmann M, Haag A, Kalinke U, Pliska V,Hengartner H, Zinkernagel R. Early high-affinity neutralizinganti-viral IgG responses without further overall improvementsof affinity. Proc Natl Acad Sci USA 1995; 92:1257–1261.

71. Barber BH. The immunotargeting approach to adjuvant-independent subunit vaccine design. Semin Immunol 1997; 9:293–301.

72. Kalinke U, Bucher E, Ernst B, Oxenius A, Roost H, Geley S,Kofler R, Zinkernagel R, Hengartner H. The role of somaticmutation in the generation of the protective humoral immuneresponse against vesicular stomatitis virus. Immunity 1996;5:639–652.

73. Kalinke U, Bucher E, Ernst B, Oxenius A, Roost H, Geley S,Kofler R, Zinkernagel R, Hengartner H. Virus neutralizationby germ-line vs. hypermutated antibodies. Proc Natl AcadSci USA 2000; 97:10126–10131.

74. Kim SH, Park SY. Selection and characterization of humanantibodies against hepatitis B virus surface antigen by phagedisplay. Hybrid Hybridomics 2002; 21:385–392.

75. Marks JD, Hoogenboom HR, Griffiths AD, Winter G. Molecularevolution of proteins on filamentous phage.Mimicking the strat-egy of the immune system. J Biol Chem 1992; 267:16007–16010.

76. Vaughan TJ, Williams AJ, Pritchard K, Osbourn JK, Pope AR,Earnshaw JC, McCafferty J, Hodits RA, Wilton J, Johnson KS.Human antibodies with sub-nanomolar affinities isolated froma large non-immunized phage display library. Nat Biotechnol1996; 14:309–314.

77. Gavilondo JV, Larrick JW. Antibody engineering at the mil-lennium. Biotechniques 2000; 29:128–145.

78. Holt L, Enever C, de Wildt R, Tomlinson I. The use of recom-binant antibodies in proteomics. Cur Opin Biotechnol 2000;11:445–449.


79. O’Connell D, Becerril B, Roy-Burman A, Daws M, Marks JD.Phage versus phagemid libraries for generation of humanmonoclonal antibodies. J Mol Biol 2002; 321:49–56.

80. Muyldermans S. Single domain camel antibodies: current sta-tus. Rev Mol Biotechnol 2001; 74:277–302.

81. Schofield DJ, Glamann J, Emerson SU, Purcell RH. Identifica-tion by phage display and characterization of two neutralizingchimpanzee monoclonal antibodies to the hepatitis E viruscapsid protein. J Virol 2000; 74:5548–5555.

82. Glamann J, Hirsch VM. Characterization of a macaque recom-binant monoclonal antibody that binds to a CD4-induced epi-tope and neutralizes simian immunodeficiency virus. J Virol2000; 74:7158–7163.

83. Yuan D, Tucker PW. Regulation of IgM and IgD synthesis in Blymphocytes. I. Changes in biosynthesis of mRNA for mu andgamma chains. J Immunol 1984; 132:1561–1565.

84. Lefkovits I. . . . and such are little lymphocytes made of. ResImmunol 1995; 146:5–10.

85. Bachmann MA, Kundig TM, Kalberer CP, Hengartner H,Zinkernagel RM. How many B cells are needed to protectagainst a virus? J Immunol 1994; 152:4235–4241.

86. Dilosa R, Maeda K, Masuda A, Szakal A, Tew J. Germinal cen-ter B cells and antibody production in the bone marrow. JImmunol 1991; 146:4071–4077.

87. Slifka MK, Matloubian M, Ahmed R. Bone marrow is a majorsite of long-term antibody production after acute viral infec-tion. J Virol 1995; 69:1895–1902.

88. Yip Y, Hawkins N, Clark M, Ward R. Immunotechnology 1997;3:195–203.

89. Thompson MA, Cancro MP. Dynamics of B cell repertoire for-mation: normal patterns of clonal turnover are altered byligand interaction. J Immunol 1982; 129:2372–2376.

90. O’Brien R, Aitken PM, O’Neil BW, Campo MS. Generation ofnative bovine mAbs by phage display. Proc Natl Acad SciUSA 1999; 96:640–645.


91. Lee W, Kohler H. Decline and spontaneous recovery of themonoclonal response to phosphorylcholine during repeatedimmunization. J Immunol 1974; 113:1644–1654.

92. Neron S, Lemieux R. Negative effect of multiple antigen injec-tions on the yield of murine monoclonal antibodies obtained byhybridoma technology. Hybridoma 1992; 11:639–644.

93. Burton DR, Barbas CF III. Antibodies from libraries. Nature1992; 359:782–783.

94. O’Brien P, Aitken R. Broadening the impact of antibody phagedisplay technology. Amplification of immunoglobulinsequences from species other than humans or mice. MethodsMol Biol 2002; 178:73–86.

95. Nuttall SD, Krishnan UV, Hattarki M, de Gori R, Irving RA,Hudson PJ. Isolation of the new antigen receptor from wobbe-gong sharks, and use as a scaffold for the display of proteinloop libraries. Mol Immunol 2001; 38:313–326.

96. Max EE. Immunoglobulin molecular genetics. In: Paul WE, ed.Fundamental Immunology. 3rd ed. New York: Raven Press,1993:315–382.

97. Tonegawa S. Somatic generation of antibody diversity. Nature1983; 302:575–581.

98. McCormack WT, Tjoelker LW, Carlson LM, Petryniak B,Barth CF, Humphries EH, Thompson CB. Chicken IgL generearrangement involves deletion of a circular episome andaddition of single nonrandom nucleotides to both coding seg-ments. Cell 1989; 56:785–791.

99. Sanz I. Multiple mechanisms participate in the generation ofdiversity of human H chain CDR3 regions. J Immunol 1991;147:1720–1729.

100. Ohlin M, Borrebaeck C. Characteristics of human antibodyrepertoires following active immune responses in vivo. MolImmunol 1996; 33:583–592.

101. Casellas R, Shih TA, Kleinewietfeld M, Rakonjac J, NemazeeD, Rajewsky K, Nussenzweig MC. Contribution of receptorediting to the antibody repertoire. Science 2001; 291:1541–1544.


102. Du Pasquier L. Phylogeny of B-cell development. Curr OpinImmunol 1993; 5:185–193.

103. Thompson CB. New insights into V(D)J recombination andits role in the evolution of the immune system. Immunity1995; 3:531–539.

104. Berek C, Brandl B, Steinhauser G. Human lambda lightchain germline genes: polymorphism in the IGVL2 genefamily. Immunogenetics 1997; 46:533–534.

105. Solin ML, Kaartinen M. Immunoglobulin constant kappagene alleles in twelve strains of mice. Immunogenetics1993; 37:401–407.

106. Ulrich HD, Moore FL, Schultz PG. Germline diversity withinthemouse Igk-V9gene family. Immunogenetics 1997; 47:91–95.

107. Solin ML, Kaartinen M. Allelic polymorphism of mouse Igh-Jlocus, which encodes immunoglobulin heavy chain joining(JH) segments. Immunogenetics 1992; 36:306–313.

108. Mattila PS, Schugk J, Wu H, Makela O. Extensive allelicsequence variation in the J region of the human immunoglobu-linheavy chain gene locus.EurJ Immunol 1995; 25:2578–2582.

109. Blankenstein T, Bonhomme F, Krawinkel U. Evolution ofpseudogenes in the immunoglobulin VH-gene family of themouse. Immunogenetics 1987; 26:237–248.

110. Adderson EE, Azmi FH, Wilson PM, Shackelford PG, CarrollWL. The human VH3b gene subfamily is highly polymorphic.J Immunol 1993; 151:800–809.

111. Milner EC, Hufnagle WO, Glas AM, Suzuki I, Alexander C.Polymorphism and utilization of human VH Genes. Ann NY Acad Sci 1995; 764:50–61.

112. Juul L, Hougs L, Barington T. A new apparently functionalIGVK gene (VkLa) present in some individuals only. Immu-nogenetics 1998; 48:40–46.

113. Flajnik MF. Comparative analyses of immunoglobulin genes:surprises and portents. Nat Rev Immunol 2002; 2:688–698.

114. Reynaud CA, Garcia C, Hein WR, Weill JC. Hypermutationgenerating the sheep immunoglobulin repertoire is an anti-gen-independent process. Cell 1995; 80:115–125.


115. Sayegh CE, Drury G, Ratcliffe MJ. Efficient antibody diver-sification by gene conversion in vivo in the absence of selec-tion for V(D)J-encoded determinants. EMBO J 1999; 18:6319–6328.

116. Reynaud CA, Anquez V, Weill JC. The chicken D locus and itscontribution to the immunoglobulin heavy chain repertoire.Eur J Immunol 1991; 21:2661.

116. Reynaud CA, Mackay CR, Muller RG, Weill JC. Somatic gen-eration of diversity in a mammalian primary lymphoid organ:the sheep ileal Peyer’s patches. Cell 1991; 64:995–1005.

117. Hein WR, Dudler L. Diversification of sheep immunoglobu-lins. Vet Immunol Immunopathol 1999; 72:17–20.

118. Lanning D, Sethupathi P, Rhee KJ, Zhai SK, Knight KL.Intestinal microflora and diversification of the rabbit anti-body repertoire. J Immunol 2000; 165:2012–2019.

119. French DL, Laskov R, Scharff MD. The role of somatic hyper-mutation in the generation of antibody diversity. Science1989; 244:1152–1157.

120. Storb U. Progress in understanding the mechanism andconsequences of somatic hypermutation. Immunol Rev1998; 162:5–11.

121. Mage RG. Diversification of rabbit VH genes by gene-conversion-like and hypermutation mechanisms. ImmunolRev 1998; 162:49–54.

122. Muramatsu M, Kinoshita K, Fagarasan S, Yamada S,Shinkai Y, Honjo T. Class switch recombination and hyper-mutation require activation-induced cytidine deaminase(AID), a potential RNA editing enzyme. Cell 2000; 102:553–563.

123. Revy P, Muto T, Levy Y, Geissmann F, Plebani A, Sanal O,Catalan N, Forveille M, Dufourcq-Labelouse R, Gennery A,Tezcan I, Ersoy F, Kayserili H, Ugazio A, Brousse N,Muramatsu M, Notarangelo L, Kinoshita K, Honjo T, FischerA, Durandy A. Activation-induced cytidine deaminase (AID)deficiency causes the autosomal recessive form of theHyper-IgM syndrome (HIGM2). Cell 2000; 102:565–575.


124. Arakawa H, Hauschild J, Buerstedde JM. Requirement of theactivation-induced deaminase (AID) gene for immunoglobu-lin gene conversion. Science 2002; 295:1301–1306.

125. Kutemeier G, Harloff C, Mocikat R. Rapid isolation of immu-noglobulin variable region genes from cell lysates of rathybridomas by polymerase chain reaction. Hybridoma 1992;11:23–31.

126. Johnson G, Wu TT. Kabat database and its applications: 30years after the first variability plot. Nucleic Acids Res 2000;28:214–218.

127. Kofler R, Geley S, Kofler H, Helmberg A. Mouse variable-region gene families: complexity, polymorphism and use innon-autoimmune responses. Immunol Rev 1992; 128:5–21.

128. Burton DR. Antibody libraries. In: Barbas CF, Burton D,Scott J, Silverman G, eds. Phage Display: A Laboratory Man-ual. Cold Spring Harbour, New York: Cold Spring HarbourLaboratory Press, 2001.

129. Andris-Widhopf J, Steinberger P, Fuller R, Rader C, BarbasCF III. Generation of antibody libraries: PCR amplificationand assembly of light- and heavy-chain coding sequences.In: Barbas CF, Burton D, Scott J, Silverman G, eds. PhageDisplay: A Laboratory Manual. Cold Spring Harbour, NewYork: Cold Spring Harbour Laboratory Press, 2001.

130. Huston JS, Levinson D, Mudgett-Hunter M, Tai MS, NovotnyJ, Margolies MN, Ridge RJ, Bruccoleri RE, Haber E, Crea R,et al. Protein engineering of antibody binding sites: recoveryof specific activity in an anti-digoxin single-chain Fv analo-gue produced in Escherichia coli. Proc Natl Acad Sci USA1988; 85(16):5879–5883.

131. Whitlow M, Bell BA, Feng SL, Filpula D, Hardman KD,Hubert SL, Rollence ML, Wood JF, Schott ME, Milenic DE.An improved linker for single-chain Fv with reduced aggrega-tion and enhanced proteolytic stability. Protein Eng 1993; 6:989–995.

132. Thiel MA, Coster DJ, Standfield SD, Brereton HM,Mavrangelos C, Zola H, Taylor S, Yusim A, Williams KA.Penetration of engineered antibody fragments into the eye.Clin Exp Immunol 2002; 128:67–74.


133. Yokota T, Milenic DE, Whitlow M, Schlom J. Rapid tumorpenetration of a single-chain Fv and comparison with otherimmunoglobulin forms. Cancer Res 1992; 52(12):3402–3408.

134. Russel M, Linderoth NA, Sali A. Filamentous phage assembly:variation on a protein export theme. Gene 1997; 192(1):23–32.


136. Enshell-Seijffers D, Smelyanski L, Gershoni JM. The rationaldesign of a ‘type 88’ genetically stable peptide display vectorin the filamentous bacteriophage fd. Nucleic Acids Res 2001;29:E50–0.

137. Rakonjac J, Model P. Roles of pIII in filamentous phageassembly. J Mol Biol 1998; 282(1):25–41.

138. Crissman JW, Smith GP. Gene-III protein of filamentousphages: evidence for a carboxyl-terminal domain with a rolein morphogenesis. Virology 1984; 132(2):445–455.

139. Reynolds C-A. Int Rev Immunol 1997; 15:265.

140. Masteller EL, Pharr GT, Funk PE, Thompson CB. Avian Bcell development. Int Rev Immunol 1997; 15:185–206.

141. Knight KL, Winstead CR. Generation of antibody diversity inrabbits. Curr Opin Immunol 1997; 9:228–232.

142. DuenasM,Vazquez J, AyalaM, SoderlindE, OhlinM, Perez L,BorrebaeckCA,Gavilondo JV. Intra- and extracellular expres-sion of an scFv antibody fragment in E. coli: effect of bacterialstrains and pathway engineering usingGroES=L chaperonins.Biotechniques 1994; 16:476–477, 480–483.

143. Rader C, Cheresch DA, Barbas CF III. A phage displayapproach for rapid antibody humanization; Designed combi-natorial V gene libraries. Proc Natl Acad Sci USA 1998;95:8910–8915.

144. Rader C, Popkov M, Neves JA, Barbas CF III. Integrin avb3targeted therapy for Kaposi’s sarcoma with an in vitroevolved antibody. FASEB J 2002; 16:2000–2002.

145. Brodeur PH, Riblet R. The immunoglobulin heavy chain vari-able region (Igh-V) locus in the mouse. I. One hundred Igh-V


genes comprise seven families of homologous genes. Eur JImmunol 1984; 14:922–930.

146. Kurosawa Y, Tonegawa S. Organization, structure, andassembly of immunoglobulin heavy chain diversity DNA seg-ments. J Exp Med 1982; 155:201–218.

147. Gough NM, Bernard O. Sequences of the joining region genesfor immunoglobulin heavy chains and their role in generationof antibody diversity. Proc Natl Acad Sci USA 1981; 78:509–513.

148. Kirschbaum T, Jaenichen R, Zachau HG. The mouse immu-noglobulin kappa locus cont6ains about 140 variable genesegments. Eur J Immunol 1996; 26:1613–1620.

149. Kabat EA, Wu TT, Reid-Miller M, Perry HM, Gottesman KS,Foeller F. Sequences of Proteins of Immunological Interest.4th ed. Washington DC: United States Department of Healthand Human Services, 1987.

150. Zhou, H, Fisher RJ, Papas TS. Optimization of primersequences for mouse scFv repertoire display library construc-tion. Nucleic Acids Res 1994; 22:888–889.

151. Dihn Q, Weng N-P, Kiso M, Ishida H, Hasegawa A, MarcusDM. High affinity antibodies against Lex and Sialyl Lex froma phage display library. J Immunol 1996; 157:732–738.

152. Kimura H, Suzuki H, Matsuzawa S. Quantitation of hemag-glutinins by the planimetry of precipitated erythrocyte pat-terns. 4. Application to Lewis blood group determinationfrom saliva and urine specimens and their stains. NipponHoigaku Zasshi 1989; 43:469–477.

152. Kimura H, Buescher ES, Ball ED, Marcus DM. Restrictedusage of VH and Vk genes by murine monoclonal antibodiesagainst 3-fucosyllactosamine. Eur J Immunol 1989; 19:1741.

153. Ball ED, Mills L, Hurd D, McMillan R, Gringrich R. Autolo-gous bone marrow transplantation for acute myeloid leuke-mia using monoclonal antibody-purged bone marrow. ProgClin Biol Res 1992; 377:97.

154. Williamson RA, Peretz D, Smorodinsky N, Bastidas R, Ser-ban H, Mehlorn I, DeArmond S, Pruisner S, Burton DR. Cir-cumventing tolerance to generate autologous monoclonal


antibodies to the prion protein. Proc Natl Acad Sci USA 7282;93:1996–7279.

155. Williamson RA, Peretz D, Pinilla C, Ball H, Bastidas R,Rozenshteyn R, Houghten R, Pruisner S, Burton DR. Map-ping the prion protein using recombinant antibodies. J Virol1998; 72:9413–9418.

156. Brocks B, Garin-Chesa P, Behrle E, Park JE, Rettig WJ,pfizenmaier K, Moosmayer D. Species-crossreactive scFvagainst the tumor stroma marker ‘‘fibroblast activationprotein’’ selected by phage display from an immunizedFAP�=– knock-out mouse. Mol Med 2001; 7:461–469.

157. Chowdhury PS, Chang K, Pastan I. Isolation of anti-mesothe-lin antibodies from a phage display library. Mol Immunol1997; 34:9–20.

158. Reiter Y, Pastan I. Antibody engineering of recombinant Fvimmunotoxins for improved targeting of cancer: disulfide-stabilized Fv immunotoxins. Clin Cancer Res 1996; 2:45–252.

159. Lorimer IAJ, Keppler-Hafkemeyer A, Beers RA, Pegram CN,Bigner DD, Pastan I. Recombinant immunotoxins specific fora mutant epidermal growth factor receptor: targeting with asingle chain antibody variable domain isolated by phage dis-play. Proc Natl Acad Sci USA 1996; 93:14815–14820.

160. Rojas G, Almagro JC, Acevedo B, Gavilondo JV. Phage anti-body fragments library combining a single human light chainvariable region with immune mouse heavy chain variableregions. J Biotechnol 2002; 23:287–298.

161. Spieker-Polet H, Sethupathi P, Yam PC, Knight KL. Rabbitmonoclonal antibodies: generating a fusion partner to pro-duce rabbit–rabbit hybridomas. Proc Natl Acad Sci USA1995; 92:9348–9352.

162. Ridder R, Schmitz R, Legay F, Gramm H. Generation of rab-bit monoclonal antibody fragments from a combinatorialphage display library and their production in the yeast Pichiapastoris. Biotechnology 1995; 13:255–260.

163. Lang IM, Barbas CF III, Schleef R. Recombinant rabbit Fabwith binding activity to type-1 plasminogen activator inhibitor


derived from a phage-display library against human alpha-granules. Gene 1996; 172:295–298.

164. Knight KL, Crane MA. . Adv Immunol 1994; 56:179–218.

165. Foti M, Granucci F, Ricciardi-Castagnoli P, Spreafico A,Ackermann M, Suter M. Rabbit monoclonal Fab derived froma phage display library. J Immunol Methods 1998; 213:201–212.

166. Rader C, Ritter G, Nathan S, Elia M, Gout I, Jungbluth AA,Cohen LS, Welt S, Old LJ, Barbas CF III. The rabbit antibodyrepertoire as a novel source for the generation therapeutichuman antibodies. J Biol Chem 2000; 275(18):13668–13676.

167. Popkov M, Mage R, Alexander C, Thundivalappil S,Barbas CF, Rader C. Rabbit immune repertoires as sourcesfor therapeutic monoclonal antibodies: the impact of kappaallotype-correlated variation in cysteine content on antibodylibraries selected by phage display. J Mol Biol 2003; 325:325–335.

168. Knight KL, Crane MA. Development of the antibody reper-toire in rabbits. Ann N Y Acad Sci 1995; 764:198–206.

169. Becker R, Knight K. Somatic diversification of immunoglobu-lin heavy chain VDJ genes: evidence for somatic gene conver-sion in rabbits. Cell 1990; 63:987–997.

170. Knight KL, Becker RS. Molecular basis of the allelic inheri-tance of rabbit immunoglobulin VH allotypes: implicationsfor the generation of antibody diversity. Cell 1990; 60:963–970.

171. Lai E, Wilson RK, Hood LE. Physical maps of the mouse andhuman immunoglobulin-like loci. Adv Immunol 1989; 46:1–59.

172. Pascual V, Verkruyse L, Casey ML, Capra JD. Analysis of IgH chain gene segment utilization in human fetal liver. Revi-siting the ‘‘proximal utilization hypothesis.’’ J Immunol 1993;151:4164–4172.

173. Sehgal D, Schiaffella E, Anderson AO, Mage RG. Generationof heterogeneous rabbit anti-DNP antibodies by gene conver-sion and hypermutation of rearranged VL and VH genes dur-ing clonal expansion of B cells in splenic germinal centers.Eur J Immunol 2000; 30:3634–3644.


174. Benammar A, Cazenave PA. A second rabbit kappa isotype. JExp Med 1982; 156:585–595.

175. Emorine L, Max EE. Structural analysis of a rabbit immuno-globulin kappa 2 J-C locus reveals multiple deletions. NucleicAcids Res 1983; 11:8877–8890.

176. Hole NJ, Harindranath N, Young-Cooper GO, Garcia R,Mage RG. Identification of enhancer sequences 30 of the rab-bit Ig kappa L chain loci. J Immunol 1991; 146:4377–4384.

177. Good PW, Notenboom R, Dubiski S, Cinader B. Basilea rabbitimmunoglobulins: detection and characterization by specificalloantiserum. J Immunol 1980; 125:1293–1297.

178. Litwin SD. Immunoglobulin allotypes. Immunol Ser 1989; 43:203–236.

179. Winstead CR, Zhai S, Sethupathi P, Knight KL. Antigen-induced somatic diversification of rabbit IgH genes: gene con-version and point mutation. J Immunol 1999; 162:6602–6612.

180. Kruithof EK, Nicolosa G, Bachmann F. Plasminogen activatorinhibitor 1: development of a radioimmunoassay and observa-tions on its plasma concentration during venous occlusion andafter platelet aggregation. Blood 1987; 70:1645–1653.

181. Chi XS, Landt Y, Crimmins DL, Dieckgraefe BK, Ladenson J.Isolation and characterization of rabbit single chain antibo-dies to human Reg I protein. J Immunol Methods 2002;266:197–207.

182. Hawlish H, Meyer zu Vilsendorf A, Bautsh W, Klos A, Kohl J.Guinea pig C3 specific rabbit single chain Fv antibodies frombone marrow, spleen and blood derived phage libraries. JImmunol Methods 2000; 236:117–131.

183. Friedman ML, Tunyaplin C, Zhai SK, Knight KL. NeonatalVH, D, and JH gene usage in rabbit B lineage cells. J Immu-nol 1994; 152:632–641.

184. Hoogenboom HR. TibTech 1997; 15:62–70.

185. Rader C, Barbas CF III. Curr Opin Biotechnol 1997; 8:503–508.

186. Hudson PJ. Curr Opin Immunol 1999; 11:548–557.


187. Steinberger P, Sutton JK, Rader C, Elia M, Barbas CF III.Generation and characterization of a recombinant humanantibody. A phage display approach for rabbit antibodyhumanization. J Biol Chem 2000; 275:36073–36078.

188. Goncalves J, Silva F, Freitas-Vieira A, Santa-Marta M,Malho R, Yang X, Gabuzda D, Barbas CF III. Functional neu-tralization of HIV-1 Vif protein by intracellular immuniza-tion inhibits reverse transcription and viral replication. JBiol Chem 2002; 277:32036–32045.

189. Larsson A, Sjoquist J. Chicken IgY: utilizing the evolutionarydifference. Comp Immunol Microbiol Infect Dis 1990; 13:199.

190. Asaoka H, Nishinaka S, Wakamiya N, Matsuda H, Murata M.Two chicken monoclonal antibodies specific for heterophilHanganutziu–Deicher antigens. ImmunolLett 1992; 32:91–96.

191. Sasai K, Lillehoj HS, Matsuda H, Wergin WP. Characteriza-tion of a chicken monoclonal antibody that recognizes the api-cal complex of Eimeria acervulina sporozoites and partiallyinhibits sporozoite invasion of CD8þ T lymphocytes in vitro.J Parasitol 1996; 82:82–87.

192. Matsushita K, Horiuchi H, Furusawa S, Horiuchi M, Shina-gawa M, Matsuda H. Chicken monoclonal antibodies againstsynthetic bovine prion protein peptide. J Vet Med Sci 1998;60:777–779.

193. Nishinaka S, Suzuki T, Matsuda H, Murata M. A new cell linefor the production of chicken monoclonal antibody by hybri-doma technology. J Immunol Methods 1991; 139:217–222.

194. Nishinaka S, Akiba H, Nakamura M, Suzuki K, Suzuki T,Tsubokura K, Horiuchi H, Furusawa S, Matsuda H. Twochicken B cell lines resistant to ouabain for the productionof chicken monoclonal antibodies. J Vet Med Sci 1996; 58:1053–1056.

195. Thompson CB, Neiman PE. Somatic diversification of thechicken immunoglobulin light chain gene is limited to therearranged variable gene segment. Cell 1987; 48:369.

196. Ratcliffe MJ, Jacobsen KA. Rearrangement of immunoglobu-lin genes in chicken B cell development. Semin Immunol1994; 6:175–184.


197. Reynaud CA, Dufour V, Weill JC. Generation of diversityin mammalian gut associated lymphoid tissues: restrictedV gene usage does not preclude complex V gene organization.J Immunol 1997; 159:3093–3095.

198. Reynaud CA, Dahan A, Anquez V, Weill JC. Somatic hyper-conversion diversivies the single VH gene of the chicken witha high incidence in the D region. Cell 1989; 59:171.

199. Arakawa H, Furusawa S, Ekino S, Yamagishi H. Immunoglo-bulin gene hyperconversion ongoing in chicken splenic germ-inal centers. EMBO J 1996; 15:2540–2546.

200. Berek C, Milstein C. Mutation drift and repertoire shift in thematuration of the immune response. Immunol Rev 1987;96:23.

201. Parvari R, Ziv E, Lantner F, Heller D, Schecher I. Somaticdiversification of chicken immunoglobulin light chains bypoint mutations. Proc Natl Acad Sci USA 1990; 87:3072.

202. McCormack WT, Thompson CB. Chicken IgL variable regiongene conversion display pseudogene donor preference and 50

to 30 polarity. Genes Dev 1990; 4:548.

203. Davies EL, Smith J, Birkett C, Manser J, Andersen-Dear D,Young J. Selection of specific phage-display antibodies usinglibraries derived from chicken immunoglobulin genes.J Immunol Methods 1995; 186:125–135.

204. Michael N, Accavitti M, Masteller E, Thompson C. The anti-gen binding characteristics of MAbs derived from in vivopriming of avian B cells. Proc Natl Acad Sci USA 1995; 95:1166–1171.

205. Yamanaka HI, Inoue T, Ikeda-Tanaka O. Chicken monoclo-nal antibody isolated by a phage display system. J Immunol1996; 157:1156–1162.

206. Cary S, Lee J, Wagenknecht R, Silverman GJ. Characteriza-tion of superantigen-induced clonal deletion with a novel clanIII-restricted avian monoclonal antibody: exploiting evolu-tionary distance to create antibodies specific for a conservedVH region surface. J Immunol 2000; 164:4730–4741.

207. Davies J, Riechmann L. Single antibody domains as smallrecognition units: design and in vitro antigen selection of


camelized, human VH domains with improved protein stabi-lity. Protein Eng 1996; 9:531–537.

208. Ghahroudi MA, Desmyter A, Wyns L, Hamers R,Muyldermans S. Selection and identification of single domainantibody fragments from camel heavy-chain antibodies.FEBS Lett 1997; 414:521–526.

209. Vu KB, Ghahroudi MA, Wyns L, Muyldermans S. Compar-ison of Ilama VH sequences from conventional and heavychain antibodies. Mol Immunol 1997; 34:1121–1131.

210. Tucker EM, Dain AR, Wright LJ, Clark SW. Culture of sheep� mouse hybridoma cells in vitro. Hybridoma 1981; 1:77.

211. Dufour V, Nau F, Malings S. The sheep Ig variable regionrepertoire consists of a single VH family. J Immunol 1996;156:2163–2170.

212. Charlton KA, Moyle S, Porter AJR, Harris WJ. Analysis of thediversity of a sheep antibody repertoire as revealed from a bac-teriophage display library. J Immunol 2000; 164:6221–6229.

213. Motyka B, Reynolds JD. Apoptosis is associated with theextensive B cell death in the sheep ileal Peyer’s patch andthe chicken bursa of Fabricius: a possible role in B cell selec-tion. Eur J Immunol 1991; 21:1951–1958.

214. Weill JC, Reynaud CA. Rearrangement=hypermutation=geneconversion: when, where and why? Immunol Today 1996; 17:92–97.

215. Hein WR, Dudler L. Diversity of Ig light chain variable regiongene expression in fetal lambs. Int Immunol 1998; 10:1251–1259.

216. Reynaud CA, Weill JC. Postrearrangement diversificationprocesses in gut-associated lymphoid tissues. Curr TopMicrobiol Immunol 1996; 212:7–15.

217. Li Y, Cockburn W, Kilpatrick J, Whitlam GC. Selection ofrabbit single-chain Fv fragments against the herbicide atra-zine using a new phage display system. Food Agric Immunol1999; 11:5.

218. Charlton KA, Harris WJ, Porter AJR. The isolation of super-sensitive anti-hapten antibodies from combinatorial antibody


libraries derived from sheep. Biosens Bioelectron 2001; 16:639–646.

218a. Jenne CN, Kennedy LJ, McCullagh P, Reynolds JD. A newmodel of sheep Ig diversification: shifting the emphasistoward combinatorial mechanisms and away from hypermu-tation. J Immunol 2003; 70:3739–3750.

219. Guidry AJ, Srikumaran S, Goldsby RA. Production and char-acterization of bovine immunoglobulins from bovine multimurine hybridomas. Methods Enzymol 1986; 121:244–265.

220. Lucier M, Thompson R, Waire J, Lin A, Osborne B, GoldsbyR. Multiple sites of V lambda diversification in cattle.J Immunol 1998; 161:5438–5444.

221. Sinclair MC, Gilchrist J, Aitken R. Bovine IgG repertoire isdominated by a single diversified VH gene family. J Immunol1997; 159:3883–3889.

222. Armour KL, Tempest PR, Fawcett PH, Fernie ML, King SI,White P, Taylor G, Harris WJ. Sequences of heavy and lightchain variable regions from four bovine immunoglobulins.Mol Immunol 1994; 31:1369–1372.

223. Berens SJ, Wylie DE, Lopez OJ. Use of a single VH family andlong CDR3s in the variable region of cattle Ig heavy chains.Int Immunol 1997; 9:189–99.

224. Parng C, Hansal S, Goldsby R, Osborne B. Gene conversioncontributes to Ig light chain diversity in cattle. J Immunol1996; 157:5478–5486.

225. Meyer A, Parng CL, Hansal SA, Osborne BA, Goldsby RA.Immunoglobulin gene diversification in cattle. Int Rev Immu-nol 1997; 15:165–183.

226. O’Brien P, Maxwell G, Campo MS. Bacterial expression andpurification of recombinant bovine Fab fragments. ProteinExpr Purif 2002; 24:43–50.

227. Hamers-Casterman C, Atarhouch T, Muyldermans S, et al.Naturally occurring antibodies devoid of light chains. Nature1993; 363:446–448.

228. van der Linden R, de Gaus B, Stok W, Bos W, van WassenaarD, Verrips T, Frenken L. Induction of immune responses and


molecular cloning of the heavy chain antibody repertoire ofLama glama. J Immunol Methods 2000; 240:185–195.

229. Muyldermans S, Atarhouch T, Saldanha J, Barbosa JARG,Hamers R. Sequence and structure of VH domain from natu-rally occurring camel heavy chain immunoglobulins lackinglight chains. Protein Eng 1994; 7:1129–1135.

230. Nguyen VK, Hamers R, Wins L, Muyldermans S. Camelheavy-chain antibodies: diverse germline VHH and specificmechanism enlarge the antigen-binding repertoire. EMBOJ 2000; 19:921–931.

231. Transue TR, de Genst E, Ghahroudi MA, Wins L,Muyldermans S. Proteins 1998; 32:515–522.

232. Frenken LGJ, van der Linden RHJ, Hermans PWJJ, Bos JW,Ruuls RC, de Geus B, Verrips CT. Isolation of antigen specificllama VHH antibody fragments and their high level secretionby Saccharomyces cerevisiae. J Biotechnol 2000; 78:11–21.

233. Lauwereys M, Ghahroudi MA, Desmyter A, Kinne J, HolzerW, De Genst E, Wyns L, Muyldermans S. Potent enzymeinhibitors derived from dromedary heavy-chain antibodies.EMBO J 1998; 17:3512–3520.

234. van Djik MA, van de Winkel JGJ. Human antibodies asnext generation therapeutics. Curr Opin Chem Biol 2001; 5:368–374.

235. Gorman JG. Rh immunoglobulin in prevention of hemolyticdisease of newborn child. NYState JMed 1968; 68:1270–1277.

236. Beasley RP, Hwang LY, Lee GC, Lan CC, Roan CH, HuangFY, Chen CL. Prevention of perinatally transmitted hepatitisB virus infections with hepatitis B virus infections with hepa-titis B immune globulin and hepatitis B vaccine. Lancet 1983;2:1099–1102.

237. Enria DA, Briggiler AM, Fernandez NJ, Levis SC, MaizteguiJI. Importance of dose of neutralising antibodies in treatmentof Argentine haemorrhagic fever with immune plasma. Lan-cet 1984; 2:255–256.

238. Buglio AF, Wheeler R, Trang J, Haynes A, Rogers K, HarveyE, Sun L, Grayeb J, Khazaeli M. Mouse=human chimeric


monoclonal antibody in man: kinetics and immune response.Proc Natl Acad Sci USA 1989; 86:4220–4224.

239. Endo T, Wright A, Morrison SL, Kobata A. Glycosylation ofthe variable region of immunoglobulin G-site specific matura-tion of the sugar chains. Mol Immunol 1995; 32:931–940.

240. Brekke O, Sandlie I. Therapeutic antibodies for human dis-eases at the dawn of the twenty first century. Nat Rev2003; 2:52062.

241. Kretzschmar T, von Ruden T. Antibody discovery: phage dis-play. Curr Opin Biotechnol 2002; 13:598–602.

242. Abrams PG, Rossio JL, Stevenson HC, Foon KA. Optimalstrategies for developing human–human monoclonal antibo-dies. Methods Enzymol 1986; 121:107.

243. Matsuda F, Ishii K, Bourvagnet P, Kuma K, Hayashida H,Miyata T, Honjo T. The complete nucleotide sequence of thehuman immunoglobulin heavy chain variable region locus.J Exp Med 1998; 188:2151–2162.

244. Meffre E, Catalan N, Seltz F, Fischer A, Nussenzweig MC,Durandy A. Somatic hypermutation shapes the antibodyrepertoire of memory B cells in humans. J Exp Med 2001;194:375–378.

245. Sasso E, Van Dijk K, Milner E. Prevalence and polymorphismof human VH3 genes. J Immunol 1990; 145:2751–2757.

246. van Dijk K, Sasso E, Milner EC. Polymorphism of the humanIg VH4 gene family. J Immunol 1991; 146:3646–3651.

247. Sasso EH, Buckner JH, Suzuki LA. Ethnic differences in VH

gene polymorphism. Ann N Y Acad Sci 1995; 764:72–73.

248. Rao SP, Huang SC, Milner EC. Analysis of the VH3 reper-toire among genetically disparate individuals. Exp ClinImmunogenet 1996; 13:131–138.

249. Coomber DWJ, Hawkins NJ, Clark MA, Ward RL. Genera-tion of anti-p-53 Fab from individuals with colorectal cancerusing phage display. J Immunol 1999; 163:2276–2283.

250. McIntosh RS, Asghar MS, Watson PF, Kemp EH, WeetmanAP. Cloning and analysis of IgGk and IgGl anti-thyroglobulin


autoantibodies from a patient with Hashimoto’s thyroiditis. JImmunol 1996; 157:927–935.

251. Clark MA. Standard protocols for the construction of Fablibraries. In: O’Brien PM, Aitken R, eds. Antibody PhageDisplay. Methods and Protocols. Totowa, New Jersey:Humana Press, 2002.

252. Winter G, Milstein C. Man-made antibodies. Nature 1991;349:293–299.

253. Green LL. Antibody engineering via genetic engineering ofthe mouse: XenoMouse strains are a vehicle for the facile gen-eration of therapeutic human monoclonal antibodies. JImmunol Methods 1999; 231:11–23.

254. Ishida I, Tomizuka K, Yoshida H, Tahara T, Takahashi N,Ohguma A, Tanaka S, Umehashi M, Maeda H, Nozaki C,Halk E, Lonberg N. Production of human monoclonal andpolyclonal antibodies in TransChromo animals. Cloning StemCells 2002; 4:91–102.

255. Mosier DE, Gulizia RJ, Baird SM, Wilson DB. Transfer of afunctional human immune system to mice with severe com-bined immunodeficiency. Nature 1988; 335:256–259.

256. Tsui P, Tornetta MA, Ames RS, Bankosky BC, Griego S,Silverman C, Porter T, Moore G, Sweet RW. Isolation of aneutralizing human RSV antibody from a dominant, non-neutralizing immune repertoire by epitope-blocked panning.J Immunol 1996; 157:772–781.

257. Chang Q, Zhong Z, Lees A, Pekna M, Pirofski L. Structure–function relationships for human antibodies to pneumococcalcapsular polysaccharide from transgenic mice with humanimmunoglobulin loci. Infect Immun 2002; 70:4977–4986.

258. Mukherjee J, Chios K, Fishwild D, Hudson D, O’Donnell S,Rich SM, Donohue-Rolfe A, Tzipori S. Production and charac-terization of protective human antibodies against Shiga toxin1. Infect Immun 2002; 70:5896–5899.

259. He Y, Honnen WJ, Krachmarov CP, Burkhart M, KaymanSC, Corvalan J, Pinter A. Efficient isolation of novel humanmonoclonal antibodies with neutralizing activity against


HIV-1 from transgenic mice expressing human Ig loci.J Immunol 2002; 169:595–605.

260. Yang X, Jia X, Corvalan J, Wang P, Davis C, Jakobovits A.Eradication of established tumors by a fully human monoclo-nal antibody to the epidermal growth factor receptor withoutconcomitant chemotherapy. Cancer Res 1999; 59:1236–1243.

261. Davis C, Gallo M, Corvalan J. Transgenic mice as a source offully human antibodies for the treatment of cancer. CancerMetastasis Rev 1999; 18:421–425.

262. Meyers K, Allen J, Gehret J, Jacobovits A, Gallo M, NeilsonE, Hopfer H, Kalluri R, Madaio M. Human antiglomerularbasement membrane autoantibody disease in XenoMouse II.Kidney Int 2002; 61:1666–1673.

263. Yang X, Corvalan J, Wang P, Roy CM, Davis C. Fully humananti-interleukin-8 monoclonal antibodies: potential therapeu-tics for the treatment of inflammatory disease states. J Leu-koc Biol 1999; 66:401–410.

264. Burton DR, Barbas CF III, Persson M, Koenig S, Chanock R,Lerner R. A large array of human monoclonal antibodies totype 1 human immunodeficiency virus from combinatoriallibraries of asymptomatic seropositive individuals. Proc NatlAcad Sci USA 1991; 88:10134–10137.

265. Barbas SM, Ditzel HJ, Salonen AM, Yang WP, Silverman GJ,Burton DR. Human autoantibody recognition of DNA. ProcNatl Acad Sci USA 1995; 92:2529–2533.

266. Burton DR. Antibodies viruses vaccines. Nat Immunol Rev2002; 2:706.

267. Chanock RM, Crowe JE Jr, Murphy BR, Burton DR. Humanmonoclonal antibody Fab fragments cloned from combinator-ial libraries: potential usefulness in prevention and=or treat-ment of major human viral diseases. Infect Agents Dis 1993;2:118–131.

268. Burton DR, Pyati J, Koduri R, Sharp SJ, Thornton GB, Par-ren PW, Sawyer LS, Hendry RM, Dunlop N, Nara PL, BarbasCF III. Efficient neutralization of primary isolates of HIV-1by a recombinant human monoclonal antibody. Science1994b; 266:1024–1027.


269. Barbas CF 3rd, Hu D, Dunlop N, Sawyer L, Cababa D,Hendry RM, Nara PL, Burton DR. In vitro evolution of a neu-tralizing human antibody to human immunodeficiency virustype 1 to enhance affinity and broaden strain cross-reactivity.Proc Natl Acad Sci USA 1994; 91:3809–3813.

270. Parren PW, Marx PA, Hessell AJ, Luckay A, Harouse J,Cheng-Mayer C, Moore JP, Burton DR. Antibody protectsmacaques against vaginal challenge with a pathogenic R5simian=human immunodeficiency virus at serum levels giv-ing complete neutralization in vitro. J Virol 2001; 75:8340–8347.

271. Mascola JR, Stiegler G, VanCott TC, Katinger H, CarpenterCB, Hanson CE, Beary H, Hayes D, Frankel SS, Birx DL,Lewis MG. Protection of macaques against vaginal transmis-sion of a pathogenic HIV-1=SIV chimeric virus by passiveinfusion of neutralizing antibodies. Nat Med 2000; 6:207–210.

272. Baba W, Liska V, Hofmann-Lehmann R, Vlasak J, Xu W,Ayehunie S, Cavacini LA, Posner MR, Katinger HE, StieglerG, Bernacky BJ, Rizvi TA, Schmidt R, Hill LR, Keeling ME,Lu Y, Wright JE, Chou TC, Ruprecht RM. Human neutraliz-ing monoclonal antibodies of the IgG1 subtype protectagainst mucosal simian–human immunodeficiency virusinfection. Nat Med 2000; 6:200–206.

273. Schmaljohn C, Cui Y, Kerby S, Pennock D, Spik K. Produc-tion and characterization of human monoclonal antibodyfragments to vaccinia virus from a phage display combinator-ial library. Virology 1999; 258:189–200.

274. Moss B. Poxviridae and their replication. In: Fields B, KnipD, Chanock R, Hirsch M, Melnick J, Monath T, Roizman B,eds. Virology Vol. 2. 2nd ed. New York: Raven Press, 1990:2079–2111.

275. Peters CJ, Sanchez A, Rollin PE, Ksiazek T, Murphy F. Filo-viridae: Marburg and ebola viruses. In: Fields D, Knipe BN,Howley P, eds. Virology. 3rd ed. New York: Raven PressInc., 1996:1161–1176.

276. Maruyama T, Rodriguez LL, Jahrling PB, Sanchez A, KhanAS, Nichol ST, Peters CJ, Parren PW, Burton DR. Ebola


virus can be effectively neutralized by antibody produced innatural human infection. J Virol 1999; 73:6024–6030.

277. Peters C, DeLuc J. An introduction. Ebola: the virus and thedisease. J Infect Dis 1999; 179(suppl 1):ix–xvi.

278. Jahrling P, Geisbert J, Swearengen J, Jaax P, Lewis T,Huggins J, Schmidt J, LeDue J, Peters CJ. Passive immuni-zation of ebola virus infected cynomologus monkeys withimmunoglobulin from hyperimmune horses. Arch Virol1999; 11(suppl):135–140.

279. Herreros J, Lalli G, Schiavo M, Schiavo G. Pathophysiologi-cal properties of clostridial neurotoxins. In: Alouf J, FreerJ, eds. The Comprehensive Sourcebook of Bacterial ProteinToxins. 2nd ed. San Diego: Academic Press, 1999:202–229.

280. Arnon SS. Clinical trial of human botulism immune globulin.In: DasGupta BR, ed. Botulinum and Tetanus Neurotoxins,Neurotransmission and Biomedical Aspects. New York: Ple-num Press, 1993:477–482.

281. Dobrindt U, Hacker J. Plasmids, phages and pathogenicityislands: lessons on the evolution of bacterial toxins. In:Alouf J, Freer J, eds. The Comprehensive Sourcebook ofBacterial Protein Toxins. 2nd ed. San Diego: Academic Press,1999:3–26.

282. Amersdorfer P, Wong C, Smith T, Chen S, Deshpande S,Sheridan R, Marks JD. Genetics and immunological compar-ison of anti-botulinum type A antibodies from immune andnon-immune human phage libraries. Vaccine 2002;20:1640–1648.

283. Foote J, Milstein C. Kinetic maturation of an immuneresponse. Nature 1991; 352:530–532.

284. Andreakos E, Taylor PC, Feldman M. Monoclonal antibodiesin immune and inflammatory diseases. Curr Opin Biotechnol2002; 13:615–620.

285. Zeidel M, Rey E, Tami J, Fischbach M, Sanz I. Genetic andfunctional characterization of human autoantibodies usingcombinatorial phage display libraries. Ann N Y Acad Sci1995; 764:559–563.


286. del Rincon I, Zeidel M, Rey E, Harley JB, James JA,Fischbach M, Sanz I. Delineation of the human systemiclupus erythematosus anti-Smith antibody response usingphage-display combinatorial libraries. J Immunol 2000; 165:7011–7016.

287. Hexham JM, Partridge LJ, Furmaniak J, Petersen VB, CollsJC, Pegg C, Rees Smith B, Burton DR. Cloning and charac-terization of TPO autoantibodies using combinatorial phagedisplay libraries. Autoimmunity 1994; 17:167–179.

288. Roben P, Barbas SM, Sandoval L, Lecerf J-M, Stollar BD,Solomon A, Silverman G. Repertoire cloning of lupus anti-DNA autoantibodies. J Clin Invest 1996; 98:2827–2837.

289. Jury K, Sohnlein P, Vogel M, Richter W. Isolation and func-tional characterization of recombinant GAD65 autoantibo-dies derived by IgG repertoire cloning from patients withtype 1 diabetes. Diabetes 2001; 50:1976–1982.

290. Graus YF, deBaets MH, Parren PW, Berrih-Aknin S, WokkeJ, van Breda Vriesman PJ, Burton DR. Human anti-nicotinicacetylcholine receptor recombinant Fab fragments isolatedfrom thymus derived phage display libraries from Myasthe-nia Gravis patients reflect predominant specificities in serumand block the action of pathogenic serum antibodies. J Immu-nol 1997; 15:1919–1929.

291. Kunicki TJ, Newman PJ. The molecular immunology ofhuman platelet proteins. Blood 1992; 80:1386–1404.

292. Clofent-Sanchez G, Lucas S, Laroche-Traineau J, Rispal P,Pellegrin JL, Nurden P, Nurden A. Autoantibodies andanti-mouse antibodies in thrombocytopenic patients asassessed by different MAIPA assays. Br J Haemotol 1996;95:53.

293. Ishida F, Gruel Y, Brojer E, Nugent DJ, Kunicki TJ. Reper-toire cloning of a human IgG inhibitor of aIIbb3 function.The OG idiotype. Mol Immunol 1995; 32:613–622.

294. Jacobin MJ, Laroche-Traineau J, Little M, Keller A, Peter K,Welschof M, Nurden A, Clofent-Sanchez G. Human IgGmonoclonal anti-alpha(IIb)beta(3)-binding fragments derivedfrom immunized donors using phage display. J Immunol2002; 168:2035–2045.


295. Eggena M, Targan SR, Iwanczyk L, Vidrich A, Gordon LK,Braun J. Phage display cloning and characterization of animmunogenetic marker (perinuclear anti-neutrophil cyto-plasmic antibody in ulcerative colitis. J Immunol 1996;156:4005–4011.

296. Lee S, Ansari AA, Cha S. Isolation of human anti-branchedchain alpha-oxo acid dehydrogenase-E2 recombinant antibo-dies by Ig repertoire cloning in idiopathic dilated cardiomyo-pathy. Mol Cells 1999; 9:25–30.

297. Reiche N, Jung A, Brabletz T, Vater T, Kirchner T, Faller G.Generation and characterization of human monoclonal scFvantibodies against Helicobacter pylori antigens. InfectImmun 2002; 70:4158–4164.

298. Pascual V, Capra JD. VH4-21, a human gene segment over-presented in the autoimmune repertoire. Arthritis Rheum1992; 35:11.

299. Carter P. Improving the efficacy of antibody-based cancertherapies. Nat Rev Cancer 2001; 1:118–129.

300. Trikha M, Yan L, Nakada MT. Monoclonal antibodies astherapeutics in oncology. Curr Opin Biotechnol 2002;13:609–614.

301. Labrecque S, Naor N, Thomson D, Matlashewski G. Analysisof the anti-p53 antibody response in cancer patients. CancerRes 1993; 53:3468–3471.

302. Pupa SM, Menard S, Andreola S, Colnaghi MI. Antibodyresponse against the cerbB-2 oncoprotein in breast carcinomapatients. Cancer Res 1993; 53:5864–5866.

303. Cai X, Garen A. Anti-melanoma antibodies from melanomapatient immunized with genetically modified autologoustumor cells: selection of specific antibodies from single-chainFv fusion phage libraries. Proc Natl Acad Sci USA 1995;92:6537–6541.

304. Kupsch J-M, Tidman NH, Kang NV, Truman H, Hamilton S,Patel N, Bishop JAN, Leigh IM, Crowe JS. Isolation ofhuman tumor-specific antibodies by selection of an antibodyphage library on melanoma cells. Clin Cancer Res 1999;5:925–931.


305. Clark MA, Hawkins NJ, Papaioannou A, Fiddes RJ, Ward RL.Isolation of human anti-c-erbB-2 Fabs from a lymph nodederived phage display library. Clin Exp Immunol 1997;109:166–174.

306. Coomber DWJ, Ward RL. Isolation of human antibodiesagainst the central DNA binding domain of p53 from an indi-vidual with colorectal cancer using antibody phage display.Clin Cancer Res 2001; 7:2802–2808.

307. Courtenay-Luck NS, Epenetos AA, Moore R, Larche M,Pectasides D, Dhokia B, Ritter MA. Development of primaryand secondary immune responses to mouse monoclonal anti-bodies used in the diagnosis and therapy of malignant neo-plasm. Cancer Res 1986; 46:6489–6493.

308. Roovers RC, van der Linden E, Zijlema H, de Bruine A,Arends J-W, Hoogenboom HR. Evidence for a bias towardintracellular antigens in the local humoral anti-tumorimmune response of a colorectal cancer patient revealed byphage display. Int J Cancer 2001; 93:832–840.

309. Mao S, Gao C, Lo C-HL, Wirsching P, Wong C-H, Janda KD.Phage-display library selection of high-affinity human single-chain antibodies to tumor-associated carbohydrate antigenssialyl LewisX and LewisX. Proc Natl Acad Sci USA 1999;96:6953–6958.

310. Lee KJ, Mao S, Sun C, Gao C, Blixt O, Arrues S, Hom LG,Kaufmann GF, Hoffman TZ, Coyle AR, Paulson J, Felding-Habermann B, Janda KD. Phage-display selection of ahuman single-chain fv antibody highly specific for melanomaand breast cancer cells using a chemoenzymatically synthe-sized Gm3—carbohydrate antigen. J Am Chem Soc 2002;124:12439–12446.

311. Dohi T, Nores G, Hakomori S. Cancer Res 1988; 48:5680–5685.

312. Zou W, Abraham M, Gilbert M, Wakarchuk WW, JenningsHJ. Glycoconjugate J 1999; 16:507–515.

313. Ehrlich PH, Moustafa ZA, Harfeldt KE, Isaacson C, OstbergL. Potential of primate monoclonal antibodies to substitutefor human antibodies: nucleotide sequence of chimpanzeeFab fragments. Hum Antibodies Hybridomas 1990; 1:23–26.


314. Logdberg L, Kaplan E, Drelich M, Harfeldt E, Gunn H,Ehrlich P, Dottavio D, Lake P, Ostberg L. Primate antibodiesto components of the human immune system. J Med Primatol1994; 23:285–297.

315. Ogata N, Ostberg L, Ehrlich PH, Wong DC, Miller RH,Purcell RH. Markedly prolonged incubation period of hepati-tis B in a chimpanzee passively immunized with a humanmonoclonal antibody to the a determinant of hepatitis B sur-face antigen. Proc Natl Acad Sci USA 1993; 90:3014–3018.

316. Glamann J, Burton DR, Parren PW, Ditzel HJ, Kent KA,Arnold C Montefiori D, Hirsch VM. Simian immunodeficiencyvirus (SIV) envelope-specific Fabs with high-level homologousneutralizing activity: recovery from a long-term-nonprogres-sor SIV-infected macaque. J Virol 1998; 72:585–592.

317. Schofield DJ, Satterfield W, Emerson SU, Purcell RH. FourChimpanzee monoclonal antibodies isolated by phage displayneutralize hepatitis A virus. Virology 2002; 292:127–136.

318. Tordsson JM, Ohlsson LG, Abrahmsen LB, Karlstrom PJ,Lando PA, Brodin TN. Phage-selected primate antibodiesfused to superantigens for immunotherapy of malignant mel-anoma. Cancer Immunol Immunother 2000; 48:691–702.

319. Roux KH, Greenberg AS, Green L, Strelets L, Avilqa D,McKinney EC, Flajnik MF. Structural analysis of the nurseshark (new) antigen receptor (NAR): molecular convergenceof NAR and unusual mammalian immunoglobulins. ProcNatl Acad Sci USA 1998; 95:11804–11809.

320. Greenberg, AS, Avila D, Hughes M, Hughes A, McKinneyEC, Flajnik MF. A new antigen receptor gene family thatundergoes rearrangement and extensive somatic deversifica-tion in sharks. Nature 1995; 374:168–173.

321. Nuttall SD, Krishnan UV, Hattarki M, de Gori R, Irving RA,Hudson PJ. Isolation of the new antigen receptor from wobbe-gong sharks, and use as a scaffold for the display of proteinloop libraries. Mol Immunol 2001; 38:313–326.

322. Jorgenson TO, Solem ST, Espelid S, Warr GW, Brandsdal BO,Smalas A. The antibody site in Atlantic salmon; phage displayand modeling of scFv with anti-hapten binding ability. DevComp Immunol 2002; 26:201–206.


323. Frenken LG, van der Linden RH, Hermans PW, Bos JW,Ruuls RC, de Geus B, Verrips CT. Isolation of antigen specificIlama VHH antibody fragments and their high level secretionby Saccharomyces cerevisiae. J Biotechnol 2000; 78:11–21.

324. Breitling F, Dubel S, Seehaus T, Klewinghaus I, Little M. Asurface expression vector for antibody screening. Gene1991; 104:147–153.

325. McCafferty J, Fitzgerald KJ, Earnshaw J, Chiswell DJ, LinkJ, Smith R, Kenten J. Selection and rapid purification of mur-ine antibody fragments that bind a transition-state analog byphage display. Appl Biochem Biotechnol 1994; 47:157–171.

326. Ward RL, Clark MA, Lees J, Hawkins NJ. Retrieval ofhuman antibodies from phage-display libraries using enzy-matic cleavage. J Immunol Methods 1996; 189:73–82.

327. Cai et al. 1995.

328. Winthrop MD, DeNardo SJ, DeNardo GL. Development of ahyperimmune anti-MUC-1 single chain antibody fragmentsphage display library for targeting breast cancer. Clin CancerRes 1999; 5(suppl 10):3088s–3094s.

329. Peipp M, Simon N, Loichinger A, Baum W, Mahr K, ZuninoSJ, Fey GH. An improved procedure for the generation ofrecombinant single-chain Fv antibody fragments reactingwith human CD13 on intact cells. J Immunol Methods2001; 251:161–176.

330. Burmester J, Spinelli S, Pugliese L, Krebber A, Honegger A,Jung S, Schimmele B, Cambillau C, Pluckthun A. Selection,characterization and x-ray structure of anti-ampicillin sin-gle-chain Fv fragments from phage-displayed murine anti-body libraries. J Mol Biol 2001; 309:671–385.

331. Rozemuller H, Chowdhury PS, Pastan I, Kreitman RJ. Isola-tion of new anti-CD30 scFvs from DNA-immunized mice byphage display and biologic activity of recombinant immuno-toxins produced by fusion with truncated pseudomonas exo-toxin. Int J Cancer 2001; 92:861–870.

332. Crowe JE Jr, Murphy BR, Chanock RM, Williamson RA,Barbas CF III, Burton DR. Recombinant human respiratorysyncytial virus (RSV) monoclonal antibody Fab is effective


therapeutically when introduced directly into the lungsof RSV-infected mice. Proc Natl Acad Sci USA 1994; 91:1386–1390.

333. Bender E, Pilkington GR, Burton DR. Human monoclonalFab fragments from a combinatorial library prepared froman individual with a low serum titer to a virus. Hum Antibo-dies Hybrid 1994; 5:3–8.

334. Zhai W, Davies J, Shang D, Chan S, Allan J. Human recom-binant single-chain antibody fragments, specific for thehypervariable region 1 of hepatitis C virus, from immunephage-display libraries. J Viral Hept 1999; 6:115–124.

335. Itoh K, Nakagomi O, Inoue K, Tada H, Suzuki T. Recombi-nant human monoclonal Fab fragments against rotavirusfrom phage display combinatorial libraries. J Biochem 1999;125:123–129.

336. De Carvalho Nicacio C, Williamson RA, Parren PW, Lundk-vist A, Burton DR, Bjorling E. Neutralizing human Fab frag-ments against measles virus recovered by phage display. JVirol 2002; 76:251–258.

337. Jakobovits A. Production of fully human antibodies by trans-genic mice. Curr Opin Biotechnol 1995; 6:561–566.

338. Kohler G, Milstein C. Continuous cultures of fused cellssecreting antibody of predefined specificity. Nature 1975;256:495–497.

339. Solem ST, Horvik I, Killie JE, Warr GW, Jorgenson TO.Diversity of immunoglobulin heavy chain in the Atlantic sal-mon (Salmo salar L.) is contributed by genes from two paral-lel IgH isoloci. Dev Comp Immunol 2001; 25:403–417.


16

Naıve Antibody Libraries fromNatural Repertoires

CLAIRE L. DOBSON, RALPH R. MINTER, andCELIA P. HART-SHORROCK

Cambridge Antibody Technology,Cambridge, U.K.

I. INTRODUCTION

Large nonimmune antibody repertoires (>1011 antibodies) arenow used routinely for the isolation of therapeutic antibodiessince they contain high affinity antibodies and are verydiverse. Indeed, subnanomolar antibodies have been isolatedfrom large nonimmune antibody libraries. These affinitiesare comparable with those of antibodies produced in a second-ary or tertiary immune response (10�8–10�10M) (1,2). The sizeof the library is not only important for affinity; it also deter-mines the success rate of selections against a large set of anti-gens (2,3). This is well illustrated by the panel of over 1000

659

antibodies, all different in amino acid sequence, that wasgenerated to the B-lymphocyte stimulator (BLyS2) protein(4). Furthermore, in some cases, antibodies with the desiredtherapeutic characteristics have been isolated directly fromnaıve antibody libraries without the need for affinity matura-tion. For example, agonistic anti-TRAIL-R1 and anti-TRAIL-R2 antibodies which are now in clinical trials for cancer havebeen isolated directly from phage display libraries (5–7). Phagedisplay is a powerful tool for generating therapeutic drugs, asillustrated by the finding that approximately one-third of allhuman antibodies currently in clinical trials are phage displayderived withmanymore antibodies in preclinical development.

This chapter gives an overview of the process of construct-ing naıve phage display libraries and discusses a broad rangeof applications, focusing on the development of antibody ther-apeutics. Finally, it addresses the potential advances in thefield and future applications.

II. CONSTRUCTION OF NAIVE LIBRARIES

II.A. Phage Display

Antibody phage display is now established as a robust alterna-tive to hybridoma technology for the isolation of monoclonalantibodies. Since the first description of the display of antibodyvariable domains on phage (8), there have beenmany antibodyphage display libraries created. The common aim of all theselibraries has been to capture a large and diverse panel of anti-bodies, thus enabling the rapid isolation of specific antibodiesto any antigen. Such libraries can be divided into three cate-gories based on the source of antibody diversity, namely naıve,immune, and synthetic libraries. The applications of immuneand synthetic libraries are covered in separate chapters.

When constructing naıve antibody libraries, the aim is tocapture the maximum number of different antibody sequencesfrom the in vivo repertoire. It is crucial that a large number ofdifferent antibody sequences are captured if the library is to beused as a ‘‘single-pot’’ resource from which antibodies to anyantigen can be isolated. Naıve antibody libraries have been

660 Dobson et al.

cloned in both the scFv format, in which the variable heavy(VH) and variable light (VL) chains are linked together to forma single protein, and the larger Fab format, in which the heavy(VHCH1) and light (VLCL) chains associate post-translation-ally (Fig. 1). The scFv format has been favored by most as itis smaller in size and therefore more amenable to large libraryconstruction (2). For the same reason, scFv libraries arethought to be more genetically stable than Fab libraries (9)and to have expression advantages, which help maintain thediversity of the library following expression on the phage sur-face (1).

To date, large, single-pot, naıve libraries have only beencloned from human repertoires although there is no reasonwhy they cannot be cloned from other animal sources. How-ever, the human libraries have the distinct advantage thatantibodies isolated from them can be used directly as thera-peutic agents with minimal risk of rejection by the patient.This contrasts with mouse antibodies and human=mousechimeric antibodies, which can elicit a human anti-mouseantibody (HAMA) response when administered to patients,reducing both the safety and efficacy of these drugs (10,11).

II.B. An Overview of B-Cell Biology andAntibody Diversity

Considering that naıve antibody libraries aim to clone the fullspectrum of antibody sequences from the human repertoire, itis important to understand the biology of the cells which pro-duce antibodies (B-cells) in order to efficiently capture the max-imum diversity. An overview of the important features of B-celldevelopment and antibody production in vivo is shown in Fig. 2.

II.B.1. B-Cells and Antibody Function

Antibody production by B-cells is tailored to neutralisingextracellular pathogens by specifically targeting them fordestruction. The specificity of antibodies is integral to theirability to only target foreign particles, such as bacteria andviruses, for destruction without causing damage to host cells.This is possible because, although there is a large antibody

Naıve Antibody Libraries from Natural Repertoires 661

pool containing over 1010 different antibodies, each B-cell onlyproduces antibodies of a single specificity and there aremechanisms in place to select for or eliminate B-cells on thebasis of the antibodies they produce (12).

Figure 1 Schematic diagram to illustrate the basic structure of animmunoglobulin G (IgG) antibody molecule and the derivation ofFab and scFv antibody fragments. (a) Whole antibody molecule,with disulphide bonds linking the different chains, showing con-stant and variable regions on the heavy and light chains. Theregions which carry out antibody functions (Fc and antigen bindingregions) are indicated. (b) Fab, with heavy and light constantdomains linked by a disulphide bond. (c) ScFv, with the linkerbetween VH and VL.

662 Dobson et al.

II.B.2. Generation of the Primary AntibodyRepertoire

The naıve repertoire of variable heavy DNA sequences isgenerated by the recombination of three gene segments,which are termed V (variable), D (diversity), and J (joining).

Figure 2 Illustration of the stages of B-cell development withreference to B-cell location and antibody status.


Analysis of the human IgH locus on chromosome 14 hasrevealed 51 functional V segments, 23 D segments, and 6 Jsegments, which can recombine to give 7038 different VHsequences (13). Further variation, known as junctional diver-sity, is introduced at the recombination sites where the V, D,and J segments are brought together. As both recombinationsites fall within VH CDR3, this concentrates the majority ofdiversity to this region and gives a primary VH repertoire ofapproximately 106 different VH sequences. A similar processoccurs in the variable light chain where lambda or kappa Vand J segments recombine to give 360 different VL sequences(13). With additional junctional diversity in the VL CDR3,the VL repertoire consists of approximately 104 differentsequences, which can combine with the 106 VH sequences togive a primary repertoire in excess of 1010 different antibodysequences (Fig. 3) (14).

II.B.3. The Naıve B-Cell Population

Following V(D)J recombination in the bone marrow, the naıveB-cells enter the peripheral B-cell pool. These naıve cells

Figure 3 Schematic diagram to illustrate the generation ofhuman antibody diversity by V(D)J recombination. (a) At the heavychain gene locus, one V, one D, and one J segment combine to give aheavy chain variable domain, with additional junctional diversity atthe recombining sites in HCDR3. (b) At the light chain gene locus,one V and one J segment combine to give a light chain variabledomain, with junctional diversity in LCDR3.

664 Dobson et al.

make up the majority of B-cells circulating in peripheral bloodat an estimated 45% of the B-cell total (15). It is at this stagethat naıve B-cells first encounter antigen and the antibodiesthey produce are of the IgM isotype, have low affinitiesfor antigen (10–5–10�6 M�1), and their sequences are notmutated from the V, D, and J gene segments encoded in thegermline. As well as being present in peripheral blood, naıveB-cells are also found in lymphoid tissues, particularly thespleen, lymph nodes, tonsils, and Peyer’s patches. However,the majority circulate in the periphery where they are mostlikely to encounter foreign antigens (16).

II.B.4. The Germinal Centre and AffinityMaturation

Naıve B-cells, which encounter antigen and become activated,with the help of T-cells, undergo affinity maturation in germ-inal centres prior to differentiation into memory cells orplasma cells. Germinal centres are found in various lymphoidtissues, particularly the spleen, lymph nodes, tonsils, andPeyer’s patches. Each germinal centre is seeded by one ora very few B-cells which proliferate rapidly and, following sti-mulation with antigen, undergo a process known as somatichypermutation. This process targets mutations to the DNAencoding the variable, but not the constant, regions of theantibody sequence. B-cells containing antibody sequenceswith improved affinity for antigen as a result of the V-regionmutations are selected, whereas low affinity or nonfunctionalcounterparts undergo apoptosis. Mutations which cause affi-nity improvements are found more frequently within CDRregions than in framework regions.

Following somatic hypermutation, B-cells which leavethe germinal centre, as memory cells or plasma cells, containantibody sequences which have a number of mutations awayfrom the germline-encoded V-gene DNA. The average muta-tion frequency is between 2% and 4%, which is equivalentto 15–30 mutations per variable region. An additional changeoccurs in the germinal centre and this is the isotype switchwhich replaces the IgM constant regions with IgG, IgA, orIgE constant regions. A large proportion of antibodies switch


from IgM to IgG, IgA, or IgE at this stage and it was oncethought that all somatically hypermutated antibodies under-went class switching. However, evidence has recently sur-faced to suggest that a large proportion of cells leaving thegerminal centre have not undergone class switching andencode somatically hypermutated IgM molecules (17).

II.B.5. Memory and Plasma Cell Populations

Memory B-cells leave the germinal centres and form a popula-tion which recirculates through the lymph and blood as long-lived cells. Of the total B-cell population in peripheral blood,between 25% and 40% are memory B-cells (15,17). It wasoriginally thought that the majority of memory cells wereeither IgG or IgA producers but recent evidence suggests thatthe proportion of IgM producing memory cells could be as largeas 40% of the circulating memory B-cell pool (17). Another keyfeature of peripheral blood memory cells is that they have 5–10-fold increased Ig mRNA levels compared with naıve cells (17).

Upon activation with antigen, memory cells proliferateand can then differentiate into antibody-secreting plasma cells.Plasma cells occur as a low percentage (0.1%) of peripheralblood lymphocytes (PBL), but occur at relatively high levels insecondary lymphoid tissues, especially spleen and bone marrow(18). Plasma cells are known to upregulate their Ig mRNAlevels 100–180-fold over the levels in resting B-cells (19,20).

II.C. Naıve Antibody Library Construction

II.C.1. Overview of the Library ConstructionProcess

Although several different strategies have been described forthe construction of large, naıve antibody libraries, there aresix stages common to all approaches, as illustrated in Fig. 4.

II.C.2. Stage 1: Isolation of mRNA from HumanB-Cells

The choice of lymphoid organ used in the isolation of humanB-cells can have significant effects on the array of antibodies

666 Dobson et al.

that are cloned into the library. To date, all published naıvelibraries have used PBL as one of the sources of mRNA(1–3,21,22). The reason for this is that the B-cell componentof PBL contains a high proportion (approximately 45%) ofnaıve B-cells (15). This naıve B-cell population offers theadvantage of being very diverse as it represents the raw out-put from V(D) J recombination in the bone marrow, that is,B-cells which have yet to encounter antigen in the blood.The diversity of the naıve population is beneficial when tryingto clone asmany antibody sequences as possible into the phagedisplay library. The disadvantage of using PBL as a source isthat the sequences have not undergone somatic hypermuta-tion and are therefore of relatively low affinity. In contrast,the memory B-cell component of PBL, which constitutes

Figure 4 The six stages involved in the construction of a naıvehuman antibody phage display library.


approximately 25–40% of PBL B-cells, have encounteredantigen and undergone clonal expansion and somatic hypermu-tation. As a result, the memory compartment encodes a lessdiverse panel of antibody sequences but they do generally havehigher affinities due to the somatic mutations introduced.

Secondary lymphoid organs such as the spleen and ton-sils, which are both centres for somatic hypermutation, havealso been used as sources of mRNA for the production of naıveantibody phage display libraries (1–3). In contrast to PBL, theB-cells in the spleen contain a high proportion of memorycells, which have already undergone selection on antigenand therefore encode a less diverse array of antibodies. How-ever, as some of these sequences will also have undergonesomatic hypermutation, the overall affinity for antigen ofspleen-derived antibodies will be higher than that of periph-eral blood B-cells.

Bone marrow-derived naıve antibody libraries (2) can beexpected to contain antibodies derived from naıve B-cells andplasma cells. As active, antibody-secreting plasma cells areknown to upregulate their Ig mRNA levels 100–180-fold overthe levels in resting B-cells (19,20), a large proportion of anti-body sequences cloned from bone marrow will be from plasmacells. This would theoretically lead to a relatively low diversityin the cloned repertoire but those antibodies present shouldhave, on average, higher affinities for antigen thannaıveB-cells.

From a practical point of view, it is crucial at this stage ofthe construction process to ensure that sufficient RNA is iso-lated to allow the eventual construction of a large library.Although time consuming, the collection of RNA from a largenumber of human donors provides obvious benefits in terms ofRNA quantity and diversity. As many as 43 human donorshave been used to construct a single, naıve library (2).

II.C.3. Stages 2 and 3: Reverse Transcriptionof mRNA to cDNA and Amplificationof VH and VL Repertoires by PCR

Following the isolation of RNA from B-cells, it is necessary toreverse transcribe the mRNA to cDNA and then amplify the

668 Dobson et al.

VH and VL regions. The choice of oligonucleotide sequencesused to prime these reactions has implications for the even-tual library diversity. At the initial stage of reverse transcrip-tion to cDNA, it is possible to use either random hexamers orantibody-specific oligonucleotides to prime the cDNA synth-esis. The use of IgM constant region primers for this initialreaction, in order to select for diverse, antibody sequencesfrom naıve IgM expressing B-cells, has been documented(1). However, it is now apparent that as many as 40% ofcirculating memory cells also express IgM (17) and these cellswill be a source of antigen-selected, somatically hypermu-tated antibody sequences. Conversely, the use of random hex-amers to prime cDNA synthesis allows all five antibodyclasses to be represented and increases the potential diversityof the final library. As such, random hexamers have been usedmore frequently in naıve library construction (2,3,22).

The choice of oligonucleotides for the amplification of VHand VL regions from cDNAhas not changed considerably sincethe first attempts at cloning large numbers of human antibodyvariable regions (23). The 31 oligonucleotide sequences citedin that paper were designed to amplify variable regions fromall known heavy and light chain gene families and are sum-marised in Fig. 5. Additional variable heavy and variablelambda light chain genes have since been discovered (24)and so the set of required oligonucleotides to amplify allhuman heavy and light chain families has expanded slightly(2). More recently, a thorough analysis of all functional germ-line V genes on the VBASE database (24) has enabled thedesign of a definitive set of primers which have been optimisedto amplify all known V genes (25).

In all cases cited above, short oligonucleotides, with noflanking regions containing restriction sites, have been usedfor the initial amplification of all V genes, so that even the poorlyrepresented templates are amplified. In order to capture maxi-mum diversity, it is also important to perform separate PCRreactions for each gene family, rather than performing PCRreactions with mixes of several primers (23). When equimolarmixtures of several primers are used, it is thought that thiscan bias the V gene representation in the eventual library (1).


II.C.4. Stage 4: Combination of VH andVL Repertoires

Up until this stage, the VH and VL repertoires have beenhandled separately. At some point in the process, it becomesnecessary to combine the heavy and light chain DNA intothe same vector. In the case of scFv libraries, the VH andVL need to be cloned either side of a short linker sequencewhich, when translated, allows the VH and VL domains toassemble into a functional conformation. For Fab libraries,although the heavy and light chains do not need to be exp-ressed as a single protein, it is usually desirable to clone theminto a single phagemid to retain linkage between genotypeand phenotype during selections.

A frequently used strategy for joining the VH and VLDNA together is two fragment PCR assembly (26). By incor-porating complementary regions at the ends of the VH andVL repertoires, the fragments can be spliced together and aDNA polymerase used to create a double-stranded joined pro-duct. A further PCR then amplifies the newly recombined VHand VL fragments prior to the final cloning step. This final

Figure 5 Illustration of the primer sites for amplification ofhuman VH and VL repertoires. The VH or VL region is amplifiedfrom the cDNA template using a panel of primers which anneal tothe 50 end of all known VH and VL regions and primers whichcan anneal at the 30 end to all known VH and VL J-regions. TheJ-region primers can be replaced with primers which anneal toregions within the constant domain.

670 Dobson et al.

PCR is important as it generates sufficient DNA to performthe multiple ligations and electroporations which are neces-sary to generate large library sizes. This method for joiningVH and VL fragments has been used to create libraries withover 1010 transformants (2).

Alternatively, the heavy and light chain repertoires canbe cloned separately into two vectors. Both vectors can be pre-pared as plasmid DNA and digested with two restrictionenzymes to excise the VH fragments from one repertoireand prepare the other plasmid, containing the VL repertoire,as an acceptor vector. The VH fragments can then be clonedinto the VL-containing acceptor vector to recombine the VHand VL repertoires. This method has also been used to gener-ate libraries containing greater than 1010 clones (3).

All these methods maximise library diversity by creatingas many new heavy and light chain combinations as possiblein vitro. As an alternative strategy, the ability of bacteria torecombine fragments of DNA in vivo has been used to createnew heavy and light chain combinations (22,27). In the morerecent of these papers, a relatively small primary repertoire of7� 107 scFv was cloned into a vector containing two differentloxP sites, one between the VH and VL fragments and the sec-ond further downstream. Phagemid containing the scFvgenes were used to infect bacteria expressing Cre recombi-nase, using a high multiplicity of infection of 200:1. Thisenabled multiple phagemid to infect a single cell and theCre recombinase activity catalysed the exchange of DNA bet-ween different phagemid to allow new VH=VL combinationsto be formed. It is estimated that the library size followingthis in vivo recombination step was 3� 1011 (22).

II.C.5. Stage 5: Quality Control of Library

Once the library construction process is complete, thereare some important quality control checks to carry out toensure the library will be capable of being used as a diversesingle-pot resource. One of the obvious initial assessments ofthe usefulness of the library is that of library size. The firstnaıve library constructed contained around 107 different


antibody sequences and with this level of diversity, the antibo-dies isolated from the library typically had affinities for anti-gen in the micromolar range (21). A breakthrough was madewith the construction of the first very large (1010) libraries,from which clones in the nanomolar range can be isolateddirectly (2). Since that point, all naıve libraries which havebeenmade have attempted to reach the same large library sizeto ensure high affinity antibodies can be isolated without theneed for further engineering to improve affinity.

In order to obtain a large library size, it is often neces-sary to include extra PCR steps, such as those used to appendrestriction sites and to amplify the repertoire following thejoining of VH and VL fragments. The result of these methodsis that the scFv DNA may have undergone up to four PCRamplifications during library construction and this introducesthe possibility of PCR errors within the antibody sequences.In some cases, clones selected from naıve antibody librariescontain up to 22 mutations away from the original germlinegene (2). As such, some libraries have been constructed usingproof-reading polymerases to reduce the number of PCRerrors (1). However, as it is virtually impossible to clone alarge repertoire of fully germline naıve antibody sequences,given the difficulty in isolating only naıve mRNA from mixedB-cell populations, it is difficult to distinguish which varia-tions from germline are due to somatic hypermutation in vivoand which are due to PCR errors in vitro.

PCR can be used to confirm that clones in the librarycontain full length VH and VL domains. In most cases,greater than 90% of transformants contain inserts of the cor-rect size which indicates that a high efficiency was achievedat the ligation stage. Some libraries have also been testedfor the expression of full-length recombinant antibody pro-tein. Protein is expressed from individual library clones andassayed by western or dot blot using a secondary antibodywhich binds to a detection handle at the C-terminus of theprotein. Despite concerns that as few as 1–10% of transfor-mants can express full-length antibody fragments (28), itseems in practice that over 50% of transformants, in bothscFv and Fab libraries, express full-length proteins (2,3).

672 Dobson et al.

Sequence analysis of libraries gives an initial indicationof the level of antibody diversity which has been cloned.Unfortunately, relatively little data are available from thelibraries constructed to date as most have either estimateddiversity by BstNI fingerprinting (21) or sequenced clonesonly after they have been selected on antigen. However, inone case, 36 library clones were sequenced and all 36 werefound to be unique (1). In this relatively small sample of thelibrary, 15 germline VH genes and eight germline VL geneswere represented. An observation that certain gene fam-ilies, such as VH3, Vk1, and Vl3, were over-represented wasthought to be a reflection of the natural bias found in humanantibody repertoires in vivo (24,29) and also to be due to theuse of pooled primers in the initial amplification (1).

Length variation in VHCDR3 regions, a source of diver-sity arising from V(D)J recombination in B-cells, offers furthersupportive evidence of the successful cloning of a diverse reper-toire. A range of VHCDR3 lengths, from 5 to 18 amino acids,was observed in the 36 library sequences analyzed (1).

Of course, the ultimate test of a naıve antibody library iswhether it can be used to isolate a large panel of specific, highaffinity antibodies to any given antigen. Some compelling evi-dence for the abilities of naıve antibody libraries in thisregard is given in Sec. III.B.1.

II.C.6. Stage 6: Rescuing the Library as Phage

The final stage in the preparation of the antibody library is toexpress the antibody fragments on the surface of phage. Alllarge libraries have been cloned into phagemid vectors to takeadvantage of the benefits of high transformation efficiency ofsmall phagemids, as opposed to large phage vectors. ScFv orFab libraries are typically cloned immediately upstream of theM13 gene III protein so that they can be expressed as a fusionprotein and incorporated into the M13 phage surface. This pro-cess, which involves the growth of the phagemid library in bac-teria followed by infection with helper phage, introduces thepossibility of over-representation of cloneswhich have a growthadvantage.These couldbe ‘‘empty’’ phagemidswhich containno


antibody fragment, clones with truncated inserts or antibodyfragments which translate rapidly as they have no rare codons,or fold rapidly due to their primary amino acid sequence. Toavoid these clones becoming over-represented, the expressionof the fusion protein is under the control of a repressible promo-ter. When the library is grown up in the presence of the repres-sor, the expression of antibody fragments is inhibited. Removalof repressor at the same time as infection with helper phagethen allows expression of the antibody fragment and incorpora-tion into the M13 phage surface.

III. APPLICATIONS OF NAIVE LIBRARIES

III.A. Antibody Therapeutics

Antibodies selected from large, nonimmunised repertoires ofscFv fragments have a multitude of applications by virtue oftheir specificity, affinity, and the relative ease with which theycan be derived. One such application is in the development oftherapeutic antibodies, which have a more rapid discovery pro-cess and cost less to develop than traditional small moleculedrugs. While traditional drugs generally require 3–5 years torefine and test in the laboratory, antibodies can take as littleas 12–18 months to progress into full clinical development.

Therapeutic antibodies can mediate their action througha variety of different mechanisms (30). Firstly, antibodies canact by blocking the function of a target antigen. This istypically achieved by preventing access of a growth factor, acytokine, or other soluble mediator by binding directly tothe soluble factor itself or to its receptor. An example of thisis the phage display derived antibody, HUMIRA2 for thetreatment of rheumatoid arthritis and Crohn’s disease (31).This antibody acts by binding and blocking the action of theproinflammatory cytokine tumor necrosis factor a (TNFa).

Another mechanism of action is Fc-mediated targetingand is especially used for the treatment of cancer. Typically,IgG1 antibodies that bind proteins expressed in a particulardisease state are used to harness the body’s own immunesystem. Antibody-dependent cellular cytotoxicity (ADCC)

674 Dobson et al.

occurs if antibody Fc regions are recognised by receptorspresent on cytotoxic cells, such as natural killer cells, macro-phages, granulocytes, and monocytes (32). Alternatively,instead of using the patient’s immune system to destroytumor cells, antibodies to the target antigen can be used todeliver radioactive isotopes or cytotoxic drugs (33–36).

Finally, antibodies can act by modulating the function ofa cell by binding to an antigen capable of transducing intra-cellular signals. For example, phage display derived agonisticantibodies to the MuSK tyrosine kinase have been isolated(37). This initial success has been extended to the isolationof agonistic antibodies to TRAIL-R1 and TRAIL-R2, whichare currently in phase I clinical trials for the potentialtreatment of cancer (5–7).

III.A.1. Generation of Antibody Therapeutics

The generation of therapeutic antibodies usually occurs intwo phases. In the first phase, naıve antibody libraries aresampled by a series of selections and screens to identifyantibodies with the desired characteristics. Generally, it ispossible to isolate antibodies with the required function andspecificity but often the affinity needs to be improved for usein therapy. This can be achieved by constructing, selecting,and screening secondary libraries, a process also known asaffinity maturation.

The following sections give an overview of the key stepsinvolved in the generation of therapeutic antibodies. Thesesteps are highlighted in the schematic in Fig. 6.

III.A.2. Selecting Therapeutic Antibodiesfrom Naıve Libraries

The selection of antibodies from phage libraries involves thesequential enrichment of specific binding phage from a largeexcess of nonbinding clones. This can be achieved by incubatingthe phage library with the target antigen. Unbound phage areremoved by washing and phage displaying scFv thatspecifically bind the antigen are eluted by disrupting thephage–antigen interaction (e.g., by applying pH gradients,


competitive elution conditions, or proteolytic reactions). Recov-ered phage are subsequently amplified by infecting E. coli andfurther rounds of selection performed as illustrated in Fig. 7.

There are many different types of possible selectionmethodologies and these are described in detail in other chap-ters. Some of the main considerations for the selection of ther-apeutic antibodies are the quality of the antigen, its purity,and methodologies that enable the selection of high affinityantibodies.

When purified target antigen is available, phage antibodyselections are often carried out with the protein directlyadsorbed onto a plastic surface such as immunotubes (Maxisorbtubes;NalgeNunc Intl., Naperville, IL)where it is noncovalentlyassociated via electrostatic andVan-der-Waals interactions (21).

Figure 6 The stages involved in the generation of therapeuticantibodies from phage antibody libraries.

676 Dobson et al.

The main disadvantage of this method is that the adsorbed pro-teinmaybeunfoldedandantibodies isolated often fail to bind thenative protein. An improvement over the direct adsorption ofproteins on plastic surfaces is their covalent coupling to platesor beads. This method of presentation enables the protein tomaintain its native fold and thus increases the chances of select-ing for antibodies binding to the native protein. The randomorientation of the protein also increases the likelihood that allepitopes will be available for binding during selection. This isof particular importance for the selection of therapeutic antibo-dies since specific epitopes must be recognised for the antibodyto alter the function of the target protein.

For therapeutic applications, it is often desirable togenerate antibodies that bind the target antigen with high

Figure 7 Illustration of the different stages in a standard phagedisplay selection cycle.


affinity. However, it is difficult to select for high affinity anti-bodies when immobilising the target protein on a solid surfacedue to rebinding and avidity effects. By performing the selec-tion in solution, conditions can be chosen to favor affinity orkinetic parameters such as off rates (38,39). For this type ofselection, the target protein is usually biotinylated and subse-quently captured using streptavidin-coated paramagneticbeads. Although affinity or off rate-driven soluble selectionscan be used to isolate antibodies from primary libraries, thesemethods are more often used to isolate high affinity antibo-dies from secondary libraries during affinity maturation (seeSec. III.A.5). The preferential selection of mutant antibodiesof higher affinities is enabled by reducing the concentrationof the target protein below the Kd of the parent clone (38,40).

Often it is not possible to obtain the target protein in apurified form. This is the case for many proteins of therapeu-tic interest (receptors, ion channels) that only retain theirfunctionality in lipid bilayers. Whole cells or plasma mem-brane preparations where the target protein is expressedcan be used instead. However, the isolation of specific antibo-dies to target proteins on cell surfaces is usually challengingdue to background binding of phage specific for nontargetproteins. Furthermore, many proteins are present on cellsat very low densities, making the selection difficult as theantigen concentration is usually much lower than the Kd ofany antibodies in the library. Depletion and=or subtractionmethods can help with the first problem, as antibodies pre-sent in the libraries that bind these nontarget antigens aredepleted (41,42). Another technique for the selection of anti-bodies binding to cell surface antigens and potentially over-coming both the problems of low antigen density andspecificity is ProxiMol� selection (43). Antibodies binding toCCR5 and blocking MIP-1a binding were generated using thisapproach (44).

In summary, selection procedures are extremely flexibleand continue to evolve to ensure that antibodies withthe desired characteristics are isolated from phage displaylibraries. An example is described in Sec. III.B.1 to supportthe theory that the choice of selection methods influences

678 Dobson et al.

output numbers and diversity. There are multiple other selec-tion methodologies not discussed here that could be useful forthe generation of therapeutic antibodies. These include in vivoselections (45,46), selection for antibodies with intracellularactivity (47,48), as well as selections for internalisation (49),and in living animals (50). Furthermore, methods have beendeveloped to purify and stabilise membrane associated pro-teins. Such methods could provide useful solutions to selectionson ‘‘difficult’’ unpurified proteins and in particular to cell sur-face molecules. For example, tagged CCR5 was purified andreconstituted on proteoliposomes before selections (51).

III.A.3. Converting Nonhuman Antibodiesto Human

The ability of phage display selections to isolate specific anti-gen binders from a large library has also been used to createa human equivalent of murine monoclonal antibodies. Inthese cases, a murine monoclonal antibody is available whichhas the desired properties of affinity and specificity but, beingmurine, would be immunogenic if used as a therapeutic. Themurine antibody can be used as a template for the selection ofhuman heavy and light chain variable regions by phage dis-play in a process known as guided selection (52). First, themurine heavy chain is paired with a library of naıvehuman light chains and selection on antigen is performed toisolate a human light chain which is capable of replacingthe murine light chain. The next stage is to pair the selectedhuman light chain with a repertoire of human heavy chainsand select on antigen to obtain a human heavy and light chaincombination which retain the original characteristics of themurine antibody (Fig. 8a). In using the murine heavy chainto guide the selection, it is possible to isolate human antibo-dies which attain, or even improve, the affinity of the originalmurine antibody and also bind the same epitope (52).

A variation on this method, known as the parallel shuffleprocedure, has also been used successfully (53). Here, themurine light chain is paired with a human heavy chainlibrary and the murine heavy chain is paired with a human


680 Dobson et al.

Figure 8 An overview of methods for performing humanizationof murine antibodies using phage display guided selections. (a)The guided selection method in which human light and heavychains are selected in two successive steps. (b) The parallel shuffleprocedure in which human light and heavy chains are selected inparallel and then paired together. (c) A guided selection strategywhich results in an antibody containing a fully human light chainand a human=murine hybrid heavy chain. (Fig. 8 Continues on nextpage.)


Figure 8 (Continued )

682 Dobson et al.

light chain library and both are selected on antigen in parallel(Fig. 8b). Selected human heavy and light chains can then bepaired together and tested for specificity and affinity.

Finally, a third method of guided selection has beendevised which takes advantage of the important role thatthe HCDR3 region plays in determining antigen binding spe-cificity. In this method, the C-terminal end of the murineheavy chain, including HCDR3 and framework 4, as well asthe murine light chain are combined with a library of humanheavy chain regions from framework 1 to framework 3. Oncea human=murine hybrid heavy chain has been selected, it ispaired with a library of human light chains and selected onantigen (Fig. 8c). The final heavy and light chain combinationare fully human, apart from the murine HCDR3 and frame-work 4 region that was used to guide the selection. Using thismethod, a human scFv was obtained which retained the epi-tope specificity of the murine equivalent (54).

The phage display derived anti-TNFa antibody,HUMIRA2, currently on the market for rheumatoid arthritis,was generated by guided selections using a mouse monoclonalantibody (31). Other guided selection derived antibodies arecurrently in preclinical development, demonstrating that, insome cases, it is a valuable alternative to the denovo generationof antibody from antibody libraries (Cambridge Antibody Tech-nology, unpublished).

III.A.4. Screening for Therapeutic Antibodies

Selections are designed to yield antibodies that bind to a par-ticular target protein. Since it is not always known whichregion of the target antigen needs to be recognised by anantibody to elicit a biological response, it is desirable to selectmultiple different antibodies to the same target that bind todifferent epitopes. To do so, it is important to analyze selectedclones after as small a number of selection rounds as possible,when the selection output is at its most diverse. The aim ofa screening strategy is to identify, within a population ofbinders, those antibodies that have desirable therapeuticcharacteristics in terms of specificity, affinity, and function.


Initially, clones are tested for target antigen specificityin an enzyme-linked immunosorbent assay (ELISA) usingunpurified phage antibodies. This is performed together withsequence analyses to identify a population of unique antigenbinders. Maximal numbers (up to 105) of sequence diverseclones are then screened to identify a subset of antibodies thatbind to a biologically relevant epitope. Such screens areusually biochemical assays that are fast, robust, automatable(e.g., 96-well or 384-well format), and use soluble antibodyfragments (scFvs or Fabs) from the bacterial supernatant orperiplasm. Biochemical assays can generally be divided intoseparation based assays (in which the reaction product isdetected after its separation from the starting material) andhomogeneous assays (in which the detection of the productdoes not require a separation step). Separation based assaysinclude ELISA based screens which are commonly used toidentify antibodies which can compete with the native ligandfor binding to a target antigen. These assays have the advan-tage that the compound has usually been separated from thereaction product at the time of detection, which minimisescompound interference effects. Furthermore, these assaytypes tend to have larger signal windows than homogeneousformats as the reaction product is the only source of signalin the assay. Homogeneous assays can take a variety of for-mats depending on the target of interest. Fluorescence reso-nance energy transfer (FRET) is the basic principle behinda number of biochemical and cell-based assays that are widelyused in HTS (55). In FRET-based assays, a fluorescent mole-cule is excited by energy at a certain wavelength, and anacceptor molecule then captures the emission energy fromthe donor fluorophore. Homogeneous assays have the advan-tage that they require fewer addition or reagent transfersteps, making them easier to automate and miniaturise.

Once a panel of unique scFvs with the desired specificityhas been isolated, functional assays are used to identify a sub-set of this population with the required biological activity.Such assays are generally cell based and can take a varietyof formats depending on the biological activity under investi-gation. Functional assays can be used to study cellular

684 Dobson et al.

responses (proliferation, chemotaxis, adhesion, apoptosis,receptor upregulation), cellular biochemistry (calcium signal-ling, kinase activation), and nuclear events (reporter genes,cell cycle status). In parallel, affinity studies are often carriedout by surface plasmon resonance to characterise the antibo-dies further. At this point, it is desirable to test the antibodiesin their final therapeutic format since differences in potencycan be observed between the scFv and IgG proteins. In mostcases, biological activity is retained and it is often possibleto obtain an improvement in potency on conversion to IgGformat due to bivalent binding.

III.A.5. Secondary Libraries for Antibody AffinityMaturation

Following the selection and screening phase of the drug dis-covery process, a panel of antibodies will have been isolatedwith the desired specificity and biological characteristics. Itis possible to isolate therapeutic grade antibodies directlyfrom a primary library without the requirement for anyimprovement in affinity. For example, antibodies to TRAIL-R1 and TRAIL-R2 for the potential treatment of cancer(5–7) are currently in early phase clinical trials. In mostcases, however, it is necessary to improve antibody affinityby a process analogous to affinity maturation. This has advan-tages for therapeutic applications; by increasing the affinity ofan antibody to its target antigen, the therapeutic dose of amAb is reduced (56), whereas its therapeutic duration isincreased. In some cases, the benefit of very high affinity anti-bodies is still a matter of debate. For example, in the treat-ment of solid tumors, mAbs of too high affinity (low pM)may be counterproductive for tumor penetration (57).

The affinity maturation of antibodies consists of introdu-cing diversity in the antibody V-genes to create a secondarylibrary which can be selected and screened for antibody var-iants with higher affinities. Methods for antibody affinitymaturation have been described in detail in other chaptersand will be summarised briefly.


Secondary libraries are generated by introducing muta-tions in the V genes of the lead antibody thereby creating alibrary of variants. Although antibody–antigen crystal struc-tures can indicate which residues should be mutated toimprove binding, atomic resolution structural data are notavailable for most antibodies.

Nondirected approacheswhereby the V genes aremutatedrandomly have been successfully used to optimise antibodieswith low starting affinities (40,58–60). Methods to introducediversity in the antibody V genes include error prone PCR(38), mutator strains of bacteria (60,61), DNA shuffling (62),and chain shuffling (59,63).

For therapeutic antibodies, CDR-directed approaches arefavored over random approaches. Antibodies isolated fromlarge antibody libraries often have low nanomolar affinitiesand CDR-directed approaches have been more successful foroptimising antibodies with such affinities. Furthermore, limit-ing the mutagenesis to the CDRs is less likely to generateimmunogenic antibodies than changes in the more conservedframework regions. Ideally, residues that modulate affinityare targeted. These residues can be determined experimen-tally by chain shuffling (64), alanine scanning of the CDRs(65), parsimonious mutagenesis (66,67), or modelling (65).Structural information of antibody–antigen complexes andstudies of the natural diversity of human antibodies createdduring the in vivo primary and secondary immune responsealso suggest key residues for targeting (68). Typically, sixCDR residues are targeted at a time and residues involvedin maintaining the CDR conformations are not altered.Approaches targeting CDRs sequentially (sequential CDRwalking) are usually preferred to parallel targeting (parallelCDR walking) since additive effects of mutations are oftenunpredictable (69). So far, the most significant gains in affinityhave been obtained by optimising CDR3 regions in the lightand heavy chain (65,69).

Antibodies currently in clinical development have beenoptimised by using a range of strategies including VL shuffle(CAT-192, Trabio2) (70), VH and VL CDR3 randomisation(CAT-213) as well as targeting CDR residues known to

686 Dobson et al.

participate in antibody–antigen interactions. Such antibodiesoften have affinities below 100 pM.

III.A.6. Therapeutic Antibody Format

ScFvs or Fabs with desired function, specificity, and affinitycan be reformatted to whole IgG1, IgG2, or IgG4 isotypesdepending on the therapeutic application. Eukaryotic expres-sion vectors for rapid one step cloning of antibody genes foreither transient or stable expression in mammalian cells havebeen described (71).

The IgG1 isotype is typically used when cell killing isrequired, as in cancer. Human IgG1 in particular can triggerthe classical complement cascade after binding to cell surfacesand promote ADCC, which is an important mediator of celllysis by the bound mAb. The IgG4 isotype does not mediateactivation of the immune system and therapeutic antibodiesin this format act mainly by blocking biological interactions.

Although the majority of antibodies in clinical develop-ment are IgGs, when effector functions are not required forthe therapeutic application, scFvs or Fabs may provide analternative. These fragments are particularly suitable fortumor targeting, as they penetrate tumors more effectivelythan do whole IgGs (72). Although the body clearance ratesof unmodified scFvs and Fabs are much higher than thoseof full-length immunoglobulins, it has been shown that half-lives can be dramatically increased by coupling polyethyleneglycol to Fabs (73).

III.A.7. Preclinical and Clinical Development

Once a therapeutic lead has been identified from the drugdiscovery process, the antibody is further characterised usinga series of in vivo models to assess the biological activity andpharmacokinetics. Preclinical activities also include thecreation of antibody-expressing mammalian cell lines. Thehighest producing cell lines are used to perform large-scalebioreactions to maximise antibody production. These provideantibody material for toxicology studies and clinical trials


and will ultimately be used to provide material for the treat-ment of patients.

III.B. Examples of Phage-Derived Antibodiesin the Clinic

All phage display derived antibodies currently in clinical devel-opment have been isolated from naıve libraries (Table 1).HUMIRA2 (adalimumab), an anti-TNFa antibody for thetreatment of rheumatoid arthritis, has become the first phagedisplay derived antibody to receive approval for marketing(Cambridge Antibody Technology=Abbott Laboratories). Thisantibody is also in clinical development for a number of otherdiseases including juvenile rheumatoid arthritis, Crohn’s dis-ease, and psoriatic arthritis.

To illustrate a typical drug discovery process using phagedisplay, a programme to isolate antibodies to the BLyS proteinis described. This formed part of a collaboration betweenCambridge Antibody Technology and Human Genome Sciences.

III.B.1. Therapeutic Antibodies to BLyS

III.B.1.1. Therapeutic Hypothesis

B-lymphocyte stimulator protein (also known as BAFF,zTNF4, TALL-1, TNFSF13B, and THANK) (74–79) is a TNFrelated cytokine that plays a critical role in the regulationof B-cell maturation and development (75–77), through bind-ing to specific receptors expressed predominantly on B-cells(74,80). Elevated levels of BLyS have been found in patientswith systemic lupus erythematosus (SLE), rheumatoid arthri-tis (81,82), and Sjogren’s syndrome (83,84). These findingssuggest that blocking the biological effects of BLyS withneutralising antibodies may be an effective approach in theamelioration or long-term remission of B-cell associated auto-immune disease. With the aim of developing a therapeuticagent for autoimmune disease, LymphoStat-B2 (HumanGenome Sciences, Inc), a human, neutralising monoclonalantibody against human BLyS was generated (4,85).

688 Dobson et al.

Table 1 Summary of Antibodies Currently in Clinical or Preclinical Development That Have Been Isolatedfrom Naıve Phage Display Libraries

Product Target Disease Partner Status

HUMIRA2 (adalimumab) TNFa Rheumatoid arthritis CAT=Abbott MarketedHUMIRA2 (adalimumab) TNFa Juvenile rheumatoid arthritis CAT=Abbott Phase IIIHUMIRA2 (adalimumab) TNFa Crohn’s disease CAT=Abbott Phase IIIHUMIRA2 (adalimumab) TNFa Psoriatic arthritis CAT=Abbott Phase IIIHUMIRA2 (adalimumab) TNFa Ankylosing spondylitis CAT=Abbott Phase IIIHUMIRA2 (adalimumab) TNFa Chronic plaque psoriasis CAT=Abbott Phase IITrabio2 (lerdelimumab) TGF-b2 Scarring post glaucoma surgery CAT Phase IIICAT-2132 (bertilimumab) Eotaxin1 Allergic disorders CAT Phase IICAT-354 IL-13 Asthma CAT PreclinicalCAT-192 (metelimumab) TGF-b1 Scleroderma CAT=Genzyme Phase IIGC-1008 TGF-b1,2,3 Idiopathic pulmonary fibrosis CAT=Genzyme PreclinicalABT-874 IL-12 Autoimmune diseases CAT=Abbott Phase IILymphoStat-B2 BLyS Systemic lupus erythematosus CAT=HGSI Phase IILymphoStat-B2 BLyS Rheumatoid arthritis CAT=HGSI Phase IIHGS-ETR1 TRAIL-R1 Certain types of cancer CAT=HGSI Phase IHGS-ETR2 TRAIL-R2 Certain types of cancer CAT=HGSI Phase IABthrax2 Anthrax protective

antigenAnthrax CAT=HGSI Phase I

Naıve

Antib

odyLib

rariesfro

mNatu

ralRep

ertoires

689

III.B.1.2. Isolation of Antibodies byPhage Display

The isolation strategy was designed with the aim ofgenerating a diverse panel of antibodies to the BLyS protein.Large, nonimmunised, human scFv phage display libraries(2), recently expanded from a total of 1010–1011 clones, wereused for all selections. Three alternative selection strategieswere adopted using purified recombinant BLyS (51kDa,homotrimer) (75). The antigen was (i) immobilised on immu-notubes; (ii) biotinylated and coupled to streptavidin-coatedplates; or (iii) biotinylated and used in soluble selection(38). Three rounds of selection were carried out and phageantibodies from the second and third round screened byELISA (22) for specific binding to BLyS and not to an unre-lated (BSA) or related (TNFa) protein. Over 7500 antibodieswere screened in this way and 2730 (36%) specifically recog-nised the BLyS antigen. DNA sequencing subsequently iden-tified 1287 sequence unique scFvs, utilising a wide range ofantibody germline sequences. Therefore, by using diverseselection strategies, the authors were able to maximise anti-body diversity, increasing the likelihood of identifying anantibody to a biologically relevant epitope. To illustrate thediversity present in the panel, the closest human germlinefor each anti-BLyS antibody was identified by aligning itsnucleotide sequence to germline VH, VL, D (diversity), andJ (joining) databases (86) using BLASTN (version 2.0.9).The heavy chain germline usage for a panel of antibodiesto a single protein antigen was very extensive, as 6 of 7VH, 22 of 27 D, and 6 of 6 JH germlines were represented.Only the VH2 subfamily, which is rarely used during animmune response, was not represented in the selected panel(Fig. 9a and b). The light chain usage, although less diversethan that observed for the heavy chain, was still considerablewith 8 of 10 Vl, 3 of 6 Vk, 3 of 7 Jl, and 4 of 6 Jk germlinesexemplified (Fig. 9c and d). Since any bias in germline usageis likely to be antigen dependent, this illustrates the impor-tance of using antibody libraries with the highest levels ofdiversity.

690 Dobson et al.

Considerable diversity was also observed within theVHCDR3 (complementarity determining region 3) sequences.These regions are nongermline encoded and there is strongevidence to suggest that they are the key determinants ofspecificity in antigen recognition (87). Five hundred andsixty-eight distinct VHCDR3s varying in length from 5 to 25amino acid residues were identified within the panel of 1287anti-BLyS antibodies, suggesting that antibodies may havebeen generated to many epitopes. This level of diversity isdesirable when generating therapeutic antibodies since, atthe start of a project, the biologically relevant epitope is notalways known.

To assess the panel of antibodies for neutralising activ-ity, each scFv was tested for the ability to inhibit BLyS bind-ing to receptors expressed on IM9 cells, a myeloma cell lineexpressing significant levels of BLyS receptors. Approxi-mately 40% of the 1287 antibodies inhibited BLyS bindingto its receptor on B-cells, and the most potent scFvs exhibitedIC50 values in the low nanomolar range. To assess the panel ofantibodies for biological activity, a subset of scFvs was refor-matted to IgG1 and tested for the ability to neutralise theactivity of human BLyS protein in a murine splenocyte invitro proliferation assay, as measured by 3H-thymidine incor-poration. Antibodies demonstrating the best inhibitory profileas full IgG molecules were hBLySmAb-1 (IC50¼ 0.05nM) andhBLySmAb-2 (IC50¼ 0.08nM). These data support findingsthat potent, high affinity antibodies can be isolated directlyfrom large nonimmune phage antibody libraries (1–3,22).

III.B.1.3. Optimisation of Lead Candidates

Optimisation of the scFv corresponding to anti-BLySantibodies hBLySmAb-1 and hBLySmAb-2 was performed toidentify antibodies with enhanced inhibitory activity. Thiswas achieved by randomising blocks of six amino acid resi-dues of the VHCDR3 domain. Large secondary libraries weregenerated for hBLySmAb-1 and hBLySmAb-2. The rando-mised libraries were then subjected to stringency selectionsto isolate antibodies with improved affinities. Phage antibo-dies with higher affinity were enriched using successive


Figure 9 Anti-BLyS antibody germline usage. (a) The humanimmunoglobulin VH, (b) D and JH, (c) VL (combined Vk andVl), and (d) JL (combined Vk and Vl) germline family geneusage. The total number of anti-BLyS antibodies (out of 1287)utilising the different germline gene segments is listed togetherwith the germline locus name. Results are shown as dendro-grams illustrating familial relationships. Solid lines indicate that

692 Dobson et al.

(Caption continued from previous page) the given germline wasobserved. Note that for 571 antibodies in the panel, the VHCDR3 domains were either too short or too novel to confidentlyassign a D gene segment (no D). (From Ref. 4, with permissionfrom Elsevier.) (Fig. 9 Continues on next page.)


Figure 9 (Continued )

694 Dobson et al.

rounds of selection by decreasing the concentration of BLyS,from 50nM down to 100pM. When selecting from a secondaryphage library, the antigen concentration is typically reducedbelow the Kd of the parent clone to allow preferential selectionof higher affinity mutants (38).

From these selections, more than 30 scFvs were identi-fied that were able to inhibit binding of biotinylated BLySto its receptors on the surface of IM9 cells with improved inhi-bitory profiles. Fig. 10 illustrates the improvements that wereobserved for the two most potent scFv from each lineage,hBLySsc-1.1 and hBLySsc-2.1, compared with their respec-tive parents.

These antibodies were reformatted as IgG1 and analyzedto confirm their specificity for the target antigen. Enzyme-linked immunosorbent assay analysis demonstrated thathBLySmAb-1 and hBLySmAb-2 were highly specific for BLySand no cross-reactivity was observed to a panel of other TNFligand family members, including APRIL, LIGHT, TNFa, Fasligand, TRAIL, and TNFb, or to IL-4 or IL-18.

In addition to fine specificity for target antigen, lead anti-bodies must also demonstrate appropriate biological activityin key in vitro assays. The top candidates were assessed inthe murine splenocyte proliferation assay and hBLySmAb1.1demonstrated the greatest inhibitory activity as IgG1. Thiswas selected as the therapeutic candidate (LymphoStat-B)and taken forward into preclinical development.

III.B.1.4. Further Characterisationof LymphoStat-B2

Having demonstrated the activity of LymphoStat-B invitro, the antibody was tested in vivo to investigate its abilityto neutralise the observable effects of exogenously adminis-tered human BLyS (0.3mg=kg for 4 consecutive days). Theseeffects include increases in murine spleen weight, increases inserum IgA levels, and increases in the population of matureB-cells in the spleen. LymphoStat-B at dosages of 1–5mg=kgcompletely prevented these BLyS-induced activities and noinhibition was observed by administration of an Ig isotype


Figure 10 Improved potency of optimised BLyS single-chain anti-bodies in the receptor inhibition assay. Purified scFv were evalu-ated for their ability to inhibit biotinylated BLyS binding to itsreceptors on IM9 cells, as measured by europium-labelled streptavi-din. (A) Comparison of hBLySsc-1 and hBLySsc-1.1 (LymphoStat-B), showing that hBLySsc-1.1 results in a 10-fold improvement inpotency compared with the parental scFv hBLySsc-1. (B) Compari-son of hBLySsc-2 and hBLySsc-2.1, showing that hBLySsc-2.1 resultsin a 20-fold improvement in potency compared with the parental scFvhBLySsc-2. The IC50 values are as follows: for hBLySsc-1, 6.3nM; forhBLySsc-1.1, 0.5nM; for hBLySsc-2, 5.86nM; and for hBLySsc-2.1,0.33nM. A four-parameter logistic model was used for curve fittingand calculation of binding parameters. Values are the mean � SEMof triplicate samples. (Reprinted from Arthritis Rheum 2003.)

696 Dobson et al.

control monoclonal antibody (human IgG1) (Fig. 11). Theseresults demonstrate that LymphoStat-B is an effective in vivoantagonist of human BLyS bioactivity.

The in vivo consequences of BLyS inhibition were evalu-ated in cynomolgus monkeys. This was considered a suitableanimal model since cynomolgus BLyS is 96% identical tohuman BLyS, and LymphoStat-B inhibits the in vitro activityof cynomolgus BLyS with similar potency as that of humanBLyS. The results of the cynomolgus study demonstrated thatLymphoStat-B is able to inhibit BLyS in vivo and that inhibi-tion results in depletion of B-cell populations after a relativelyshort course of treatment.

LymphoStat-B possesses many properties that make itideally suited for use as a therapeutic agent. First, it bindswith exquisite specificity to BLyS and does not recogniseother TNF family members, including its closest homologuehuman APRIL. Second, it binds with high affinity to its targetantigen (human BLyS) and potently inhibits its activitiesboth in vitro and in vivo. As an antibody, an additional advan-tage of LymphoStat-B is its long terminal half-life. In miceand monkeys, the half-life of LymphoStat-B is 2.5 and 11–14days, respectively. This has advantages for therapeuticapplications since the therapeutic dose of the mAb is reduced,whereas its therapeutic duration is increased.

LymphoStat-B is being developed by Human GenomeSciences as a novel treatment for autoimmune diseases andis currently being evaluated in phase II clinical trials inpatients with SLE and RA. Studies are also underway to iden-tify additional diseases in which BLyS might play a role andLymphoStat-B may have a therapeutic potential.

III.C. Further Applications of Naıve Librariesand Future Directions

Antibodies isolated from large, human scFv repertoires havea broad range of applications. Besides offering a source ofantibodies for therapy, phage display technology is very wellsuited for high-throughput generation of antibodies forresearch purposes, such as in the emerging field of proteomics.


Figure 11 Inhibitory effects of LymphoStat-B on the responses of BALB=c mice stimulated with humanB-lymphocyte stimulator (BLyS). BLyS was administered to mice over a 5-day period, with or without Lympho-Stat-B or control IgGl on days 1 and 3. The effects of BLyS on spleen weight, splenic B-cell representation, andserum IgA levels were determined on day 5. (A) Effect of LymphoStat-B on BLyS-induced increases in spleenweight. (B) Effect of LymphoStat-B on BLyS-induced increases in the number of mature splenic B cells(ThBþ =B220þ). These markers are the murine equivalents of Ly6D and CD45R, respectively. Data are reportedas themean of the B220þ =ThBþ cell population, as determined by flow cytometry analysis. (C) Effect of Lympho-Stat-B on BLyS-induced elevations of serum IgA levels. Values are the mean � SEM (n¼ 10 mice per treatmentgroup). ### P < 0.0005 for recombinant humanBLyS vs. buffer; � P < 0.05 for LymphoStat-B vs. the correspond-ing dose of human IgGl; �� P < 0.0005 for LymphoStat-B vs. the corresponding dose of human IgGl; # P < 0.05for recombinant human BLyS vs. buffer. hu ¼ human; Ab ¼ antibody. (Reprinted from Arthritis Rheum 2003.)

698

Dobsonet

al.

The combination of protein chips (88) and recombinant antibo-dies selected by phage display could enable the determinationof the protein profile for any tissue or even whole organism.This could provide a useful tool for the identification of diseasespecific antigens. Although the utilisation of antibody librariesfor target discovery has not yet been explored to its full poten-tial, studies based on synthetic (42) naıve (89), and immunerepertoires have proven the concept. In contrast to hybridomatechnology, antibody libraries can be positively and negativelyselected against human cell lines, allowing specificity tobecome part of the selection. Another promising strategy, inparticular for the discovery of targets that are expressed bythe tumor vasculature, is the selection of antibody librariesin vivo. Johns et al. (50) utilised an in vivo selection process;namely intravenously injecting an antibody phage library intomice. scFv were obtained that bound specifically to thymicvascular endothelium and perivascular epithelium.

Anti-idiotypic antibodies are potentially useful as substi-tutes of antigens (90). If anti-idiotypic antibodies bind to theantigen-combining site (paratope), they are displaying an‘‘internal image’’ of the idiotypic antibody (Ab1). Such antibo-dies can potentially serve as a vaccine to induce an immuneresponse to a pathogenic antigen, thus avoiding imm-unisation with the pathogen itself. Goletz et al. (91) haveefficiently selected anti-idiotypic antibodies from naıve phagelibraries by specific elution (in combination with trypsintreatment) of the idiotype-bound with the soluble antigen asdemonstrated with two carbohydrates and one conforma-tional peptide epitope. Anti-idiotype antibodies could alsopotentially be used for the assessment of safety and pharma-cokinetic profiles of antibodies administered clinically.

Phage display is an established technology for creatinghuman antibodies and has been successfully used to generatetherapeutic antibodies. Alternative display technologies arenow emerging, including ribosome display, and covalent dis-play. Libraries with a diversity of 1014—significantly largerthan phage display libraries—are possible using cell-freesystems. Using this technique, scFvs specific for a yeast tran-scription factorGCN4have been selected. In addition, ribosome


displayhasbeen successfullyused togenerate antiprogesteroneantibody fragments. Ribosome display (acquired and developedfor commercial purposes by Cambridge Antibody Technology)has the potential to become a versatile tool for the developmentof human monoclonal antibodies with picomolar affinities.

IV. Summary

This chapter gives an overview of the process of constructingnaıve phage display libraries and discusses a range of applica-tions, focusing on the development of antibody therapeutics.All phage display derived antibodies currently in clinicaldevelopment have been isolated from naıve libraries, withHUMIRA2 becoming the first phage display derived humanantibody to receive approval for marketing. Other applica-tions for naıve libraries are now emerging. For example, theselibraries may be used for drug target identification, the devel-opment of anti-idiotypes and for the high throughput genera-tion of antibodies for research purposes. Phage display isclearly a powerful tool and will continue to be a major techni-que for the generation of antibodies in the future.

REFERENCES

1. Sheets MD, et al. Efficient construction of a large nonimmunephage antibody library: the production of high-affinity humansingle-chain antibodies to protein antigens. Proc Natl Acad SciUSA 1998; 95(11):6157–6162.

2. Vaughan TJ, et al. Human antibodies with sub-nanomolaraffinities isolated from a large non-immunized phage displaylibrary. Nat Biotechnol 1996; 14(3):309–314.

3. de Haard HJ, et al. A large non-immunized human Fabfragment phage library that permits rapid isolation and kineticanalysis of high affinity antibodies. J Biol Chem 1999; 274(26):18218–18230.

4. Edwards BM, et al. The remarkable flexibility of the humanantibody repertoire; isolation of over one thousand different

700 Dobson et al.

antibodies to a single protein, BLyS. J Mol Biol 2003; 334(1):103–118.

5. Salcedo T, et al. TRM-1, a fully humanTRAIL-R1 agonisticmono-clonal antibody, displays in vitro and in vivo antitumor activity.93rd Annual Meeting AACR, San Francisco, CA, USA, 2002.

6. Dobson CL, et al. Generation of human therapeutic anti-TRAIL-R1 agonistic antibodies by phage display. 93rd AnnualMeeting AACR, San Francisco, CA, USA, 2002.

7. Edwards BM, et al. Isolation of agonistic human monoclonalantibodies to TRAIL-R2 that display potent in vitro and in vivoanti-tumour activities. Antibody-Based Therapeutics for Can-cer, Banff, Alberta, Canada, 2003.

8. McCafferty J, et al. Phage antibodies: filamentous phagedisplaying antibody variable domains. Nature 1990; 348(6301):552–554.

9. Griffiths AD, Duncan AR. Strategies for selection of antibodiesby phage display. Curr Opin Biotechnol 1998; 9(1):102–108.

10. LoBuglio AF, et al. Mouse=human chimeric monoclonal anti-body in man: kinetics and immune response. Proc Natl AcadSci USA 1989; 86(11):4220–4224.

11. Kuus Reichel K, et al. Will immunogenicity limit the use,efficacy, and future development of therapeutic monoclonalantibodies? Clin Diagn Lab Immunol 1994; 1(4):365–372.

12. Rajewsky K. Clonal selection and learning in the antibody sys-tem. Nature 1996; 381(6585):751–758.

13. Lefranc M-P, LeFranc G. The Immunoglobulin Facts Book.Academic Press, 2001.

14. Winter G, Milstein C. Man-made antibodies. Nature 1991;349(6307):293–299.

15. Klein U, Rajewsky K, Kuppers R. Human immunoglobulin(Ig)MþIgDþ peripheral blood B cells expressing the CD27 cellsurface antigen carry somatically mutated variable regiongenes: CD27 as a general marker for somatically mutated(memory) B cells. J Exp Med 1998; 188(9):1679–1689.

16. Roitt I, Brostoff J, Male DK. Immunology. 4th ed. Mosby, 1996.


17. Klein U, Kuppers R, Rajewsky K. Evidence for a large com-partment of IgM-expressing memory B cells in humans. Blood1997; 89(4):1288–1298.

18. McHeyzer Williams MG, Ahmed R. B cell memory and the long-lived plasma cell. Curr Opin Immunol 1999; 11(2):172–179.

19. Kelley DE, Perry RP. Transcriptional and posttranscriptionalcontrol of immunoglobulin mRNA production during B lympho-cyte development. Nucleic Acids Res (Online) 1986; 14(13):5431–5447.

20. Matthes T, Kindler V, Zubler RH. Semiquantitative, nonra-dioactive RT-PCR detection of immunoglobulin mRNA inhuman B cells and plasma cells. DNA Cell Biol 1994; 13(4):429–436.

21. Marks JD, et al. By-passing immunization: human antibodiesfrom V-gene libraries displayed on phage. J Mol Biol 1991;222:581–597.

22. Sblattero D, Bradbury A. Exploiting recombination in singlebacteria to make large phage antibody libraries. Nat Biotech-nol 2000; 18(1):75–80.

23. Marks JD, et al. Oligonucleotide primers for polymerase chainreaction amplification of human immunoglobulin variablegenes and design of family-specific oligonucleotide probes.Eur J Immunol 1991; 21(4):985–991.

24. Tomlinson IM, et al. The repertoire of human germline VHsequences reveals about fifty groups of VH segments with dif-ferent hypervariable loops. J Mol Biol 1992; 227(3):776–798.

25. Sblattero D, Bradbury A. A definitive set of oligonucleotide pri-mers for amplifying human V regions. Immunotechnology1998; 3(4):271–278.

26. Horton RM, et al. Engineering hybrid genes without the use ofrestriction enzymes: gene splicing by overlap extension. Gene1989; 77(1):61–68.

27. Waterhouse P, et al. Combinatorial infection and in vivorecombination: a strategy for making large phage antibodyrepertoires. Nucleic Acids Res (Online) 1993; 21(9):2265–2266.

702 Dobson et al.

28. McCafferty J. Phage display: factors affecting panning effi-ciency. In: Kay B, Winter L, McCafferty J, eds. Display of Pep-tides and Proteins. San Diego: Academic, 1996:261–276.

29. Griffiths AD, et al. Isolation of high affinity human antibodiesdirectly from large synthetic repertoires. EMBO J 1994; 13(14):3245–3260.

30. Glennie MJ, Johnson PWM. Clinical trials of antibody therapy.Immunol Today 2000; 21(8):403–410.

31. Weinblatt ME, et al. Adalimumab, a fully human anti-tumornecrosis factor alpha monoclonal antibody, for the treatmentof rheumatoid arthritis in patients taking concomitant metho-trexate: the ARMADA trial. Arthritis Rheum 2003; 48:35–45.

32. Breedveld FC. Therapeuticmonoclonal antibodies. Lancet 2000;355(9205):735–740.

33. Esteva FJ, Hayes DF. Monoclonal antibody-based therapy ofbreast cancer. In: Grossbard ML, ed. Monoclonal Antibodty-Based Therapy of Breast Cancer. New York: Marcel Dekker,1998:309–338.

34. Bodey B, et al. Genetically engineered monoclonal antibodiesfor direct anti-neoplastic treatment and cancer cell specificdelivery of chemotherapeutic agents. Curr Pharm Des 2000;6:261–276.

35. Blattler WA, Lambert JM. Preclinical immunotoxin develop-ment. In: Grossbard ML, ed. Monoclonal Antibody Based Ther-apy of Cancer. New York: Marcel Dekker, 1998:1–22.

36. Grossbard ML, et al. Monoclonal antibody-based therapies ofleukemia and lymphoma. Blood 1992; 80:863–878.

37. Xie M, et al. Direct demonstration of MuSK involvement inacetylcholine receptor clustering through identification of ago-nist scFv. Nat Biotechnol 1997; 15:768–771.

38. Hawkins RE, Russell SJ, Winter G. Selection of phage antibo-dies by binding affinity. Mimicking affinity maturation. J MolBiol 1992; 226:889–896.

39. Duenas M, et al. Selection of phage displayed antibodies basedon kinetic constants. Mol Immunol 1996; 33(3):279–285.


40. Schier R, et al. Isolation of high affinity monomeric humananti-c-erbB-2 single-chain Fv using affinity driven selection.J Mol Biol 1996; 255:28–43.

41. Siegel DL, et al. Isolation of cell surface-specific humanmonoclonal antibodies using phage display and magnetically-activated cell sorting: applications in immunohematology.J Immunol Methods 1997; 206:73–85.

42. de Kruif J, et al. Rapid selection of cell subpopulation-specifichuman monoclonal antibodies from a synthetic phage antibodylibrary. Proc Natl Acad Sci USA 1995; 92(9):3938–3942.

43. Osbourn JK, et al. Pathfinder selection: in situ isolation ofnovel antibodies. Immunotechnology 1998; 3:293–302.

44. Osbourn JK, et al. Directed selection of MIP-1 alpha neutraliz-ing CCR5 antibodies from a phage display human antibodylibrary. Nat Biotechnol 1998; 16(8):778–781.

45. Pasqualini R, Koivunen E, Ruoslahti E. av Integrins as recep-tors for tumor targeting by circulating ligands. Nat Biotechnol1997; 15:542–546.

46. Arap W, et al. Steps toward mapping the human vasculatureby phage display. Nat Med 2002; 8(2):121–127.

47. Gargano N, Cattaneo A. Rescue of a neutralizing anti-viralantibody fragment from an intracellular polyclonal reper-toire expressed in mammalian cells. FEBS Lett 1997; 414:537–540.

48. Visintin M, et al. The intracellular antibody capture technol-ogy (IACT): towards a consessus sequence for intracellularantibodies. J Mol Biol 2002; 317:73–83.

49. Becerril B, Poul MA, Marks JD. Toward selection of internaliz-ing antibodies from phage libraries. Biochem Biophys ResCommun 1999; 255(2):386–393.

50. Johns M, George AJ, Ritter MA. In vivo selection of scFvfrom phage display libraries. J Immunol Methods 2000; 239:137–151.

51. Mirzabekov T, et al., Paramagnetic proteoliposomes contain-ing a pure, native, and oriented seven-transmembraneseg-ment protein, CCR5. Nat Biotechnol 2000; 18(6):649–654.

704 Dobson et al.

52. Jespers LS, et al., Guiding the selection of human antibodiesfrom phage display repertoires to a single epitope of an anti-gen. Biotechnology 1994; 12:899–903.

53. Wang Z, et al. Humanization of a mouse monoclonal antibodyneutralizing TNF-alpha by guided selection. J ImmunolMethods 2000; 241(1–2):171–184.

54. Klimka A, et al. Human anti-CD30 recombinant antibodies byguided phage antibody selection using cell panning. Br JCancer 2000; 83(2):252–260.

55. Selvin PR. Principles and biophysical applications of lantha-nide-based probes. Annu Rev Biophys Biomol Struct 2002;31:275–302.

56. Barbas CF III, Burton DR. Selection and evolution of high-affinity human anti-viral antibodies. Trends Biotechnol 1996;14:230–234.

57. Adams GP, Schier R. Generating improved single-chain Fvmolecules for tumor targeting. J Immunol Methods 1999; 231:249–260.

58. Deng SJ, et al. Selection of antibody single-chain variable frag-ments with improved carbohydrate binding by phage display.J Biol Chem 1994; 269:9533–9538.

59. Marks JD, et al. By-passing immunization: building high affi-nity human antibodies by chain shuffling. Biotechnology 1992;10:779–783.

60. Low NM, Holliger PH, Winter G. Mimicking somatichypermutation: affinity maturation of antibodies displayedon bacteriophage using a bacterial mutator strain. J Mol Biol1996; 260:359–368.

61. Irving RA, Kortt AA, Hudson PJ. Affinity maturation of recom-binant antibodies using E. coli mutator cells. Immunotechnol-ogy 1996; 2:127–143.

62. Crameri A, Cwirla S, Stemmer WP. Construction and evolu-tion of antibody-phage libraries by DNA shuffling. Nat Med1996; 2:100–102.

63. Clackson T, et al. Making antibody fragments using phage dis-play libraries. Nature 1991; 352:624–628.


64. Thompson J, et al. Affinity maturation of a high-affinityhuman monoclonal antibody against the third hypervariableloop of human immunodeficiency virus: use of phage displayto improve affinity and broaden strain reactivity. J Mol Biol1996; 256:77–88.

65. Schier R, et al. Isolation of picomolar affinity anti-c-erbB-2single-chain Fv by molecular evolution of the complementaritydetermining regions in the center of the antibody binding site.J Mol Biol 1996; 263:551–567.

66. Schier R, et al. Identification of functional and structuralamino-acid residues by parsimonious mutagenesis. Gene 1996;169:147–155.

67. Balint RF, Larrick JW. Antibody engineering by parsimoniousmutagenesis. Gene 1993; 137:109–118.

68. Tomlinson I, et al. The imprint of somatic hypermutation onthe repertoire of human germline V genes. J Mol Biol 1996;256(5):813–817.

69. Yang WP, et al. CDR walking mutagenesis for the affinitymaturation of a potent human anti-HIV-1 antibody into thepicomolar range. J Mol Biol 1995; 254(3):392–403.

70. Thompson J, et al. A fully human antibody neutralising biolo-gically active TGFb2 for use in therapy. J Immunol Methods1999; 227:17–29.

71. Persic L, et al. An integrated vector system for the eukaryoticexpression of antibodies or their fragments after selection fromphage display libraries. Gene 1997; 187:9–18.

72. Chester KA, Hawkins RE. Clinical issues in antibody design.Trends Biotechnol 1995; 13(8):294–300.

73. Weir ANC, et al. Formatting antibody fragments to mediatespecific therapeutic functions. Biochem Soc Trans 2002; 30(4):512–516.

74. Gross JA, et al. TACI and BCMA are receptors for a TNFhomologue implicated in B-cell autoimmune disease. Nature2000; 404:995–999.

706 Dobson et al.

75. Moore PA, et al. BLyS: member of the tumour necrosis factorfamily and B lymphocyte stimulator. Science 1999; 285(5425):260–263.

76. Schneider P, et al. BAFF, a novel ligand of the tumour necrosisfactor family, stimulates B cell growth. J Exp Med 1999;189:1747–1756.

77. Shu HB, Hu WH, Johnson H. TALL-1 is a novel member of theTNF family that is down-regulated by mitogens. J Leukoc Biol1999; 65:680–683.

78. Mukhopadhyay A, et al. Identification and characterization ofa novel cytokine, THANK, a TNF homologue that activatesapoptosis, nuclear factor-kB, and c-Jun NH2-terminal kinase.J Biol Chem 1999; 274:15978–15981.

79. Tribouley C, et al. Characterization of a new member of theTNF family expressed on antigen presenting cells. Biol Chem1999; 380(12):1443–1447.

80. Thompson JS, et al. BAFF-R, a newly identified TNF receptorthat specifically interacts with BAFF. Science 2001; 293(5537):2108–2111.

81. Zhang J, et al. Cutting edge: a role for B lymphocyte stimula-tor in systemic lupus erythematosus. J Immunol Methods2001; 166(1):6–10.

82. Cheema GS, et al, Elevated serum B lymphocyte stimulatorlevels in patients with systemic immune-based rheumaticdiseases. Arthritis Rheum 2001; 44:1313–1319.

83. Mariette X, et al. The level of BLyS (BAFF) correlates with thetitre of autoantibodies in human Sjogren’s syndrome. AnnRheum Dis 2003; 62:168–171.

84. Groom J, et al. Association of BAFF=BLyS overexpression andaltered B cell differentiation with Sjogren’s syndrome. J ClinInvest 2002; 109(1):59–68.

85. Baker KP, et al. Generation and characterization of Lympho-Stat-B, a human monoclonal antibody that antagonizes thebioactivities of B lymphocyte stimulator. Arthritis Rheum2003; 48(11):3253–3265.


86. Dilks TJ, Hardman CH, Brocklehurst SM. CAST2 Database.Cambridge Antibody Technology.

87. Xu JL, Davis MM. Diversity in the CDR3 region of V(H) is suf-ficient for most santibody specificities. Immunity 2000;13(1):37–45.

88. Haab BB, Dunham MJ, Brown PO. Protein microarrays forhighly parallel detection and quantitation of specific proteinsand antibodies in complex solutions. Genome Biol 2001;2:research 0004.1–0004.12.

89. Ridgway JB, et al. Identification of a human anti-CD55 single-chain Fv by subtractive panning of a phage library using tumorand nontumor cell lines. Cancer Res 1999; 59(11):2718–2723.

90. Jerne NK. Towards a network theory of the immune system.Ann Immunol 1974; 125C:373–389.

91. Goletz S, et al. Selection of large diversities of antiidiotypicantibody fragments by phage display. J Mol Biol 2002;315:1087–1097.

708 Dobson et al.

17

Synthetic Antibody Libraries

FREDERIC A. FELLOUSE andSACHDEV S. SIDHU

Department of Protein Engineering, Genentech, Inc.,South San Francisco, California, U.S.A.

I. INTRODUCTION

Phage display has proven to be an ideal technology for use inantibody engineering. The method has been used for affinitymaturation of natural antibodies, and also, for humanizationof murine antibodies (Chapter 14). In addition, large librariesfrom immunized donors (Chapter 15) or from naıve naturalrepertoires (Chapter 16) have been used as valuable resourcesfor antibodies with novel functions. However, while theseapplications use phage display for in vitro selection and opti-mization, they nonetheless rely on a natural source of anti-body diversity. A particularly promising branch of antibodyengineering research has focused on the design and applica-tion of synthetic antibody libraries, that is, libraries in which

709

the diversity is derived from man-made sources, rather thanfrom natural repertoires.

Libraries of this type would have numerous advantagesfrom both theoretical and practical viewpoints. In practicalterms, the use of synthetic repertoires should greatly expandthe diversities accessible by phage display technology, sincelibraries would no longer be limited to the scope of the naturalimmune system. In particular, synthetic libraries would becompletely naıve, and would not be subjected to the restric-tions imposed by self-tolerance of natural repertoires. Thus,it should be relatively simple to obtain antibodies againsteven highly conserved proteins for which conventional hybri-doma technologies are relatively ineffective. In addition, syn-thetic libraries permit the use of any framework of choice, andframeworks can be chosen for properties such as stability,high expression or lack of immunogenicity. This is particu-larly valuable for the development of therapeutic antibodies,as it has become clear that the use of non-immunogenichuman frameworks is a key requisite for the success of suchtherapies. From a more fundamental point of view, the abilityto precisely control the design of synthetic diversity offersunparalleled opportunities for the exploration of the funda-mental principles governing the structure and function ofantibodies. Synthetic libraries can be used to not only obtainantibodies with novel specificities and functions, but also,libraries can be designed to specifically address fundamentalquestions regarding the mechanisms of antigen recognition.

Despite the enormous potential of synthetic antibodylibraries, the field has developed slowly in comparison withphage display applications that use natural repertoires. Forthe most part, this has been due to the fact that the develop-ment of useful synthetic repertoires is considerably moredifficult than the simple exploitation of natural repertoires.In the latter case, libraries can be constructed from naturalsources with only aminimal understanding of the fundamentalprinciples of antibody structure and function, because the onlytechnical challenge resides in transferring the functional, invivo repertoire to an in vitro display system with molecularbiology techniques. In contrast, the development of synthetic

710 Fellouse and Sidhu

repertoires amounts to building antibody diversity fromscratch, and this requires not only molecular biology, but also,detailed and extensive knowledge of the antibody molecule.

Despite the difficulties inherent in the ab initio design ofsynthetic antibody repertoires, significant progress has beenmade in recent years. Most of the research in the field hasinvolved four major groups—the Scripps Research Institute,the Medical Research Council, Morphosys AG, and GenentechInc. In this chapter, we review the major efforts and findingsfrom these centers.

II. THE SCRIPPS RESEARCH INSTITUTE

Early work with synthetic antibody libraries at the ScrippsResearch Institute (La Jolla, CA) made use of a single antigen-binding fragment (Fab) as the framework. A simple librarywas constructed by replacing the DNA encoding for the thirdcomplementarity determining region of the heavy chain(CDR-H3) by a completely random stretch of 16 degeneratecodons encoding for all 20 natural amino acids (1). CDR-H3was chosen for diversification because it lies at the center ofthe antigen-binding site, it is the most diverse CDR in naturalrepertoires, and it often makes extensive contact with antigen(2–5). The resulting library was relatively modest in size(5� 107 clones) and was limited in terms of both frameworkand CDR diversity. Nonetheless, the library was used success-fully to obtain antibodies with novel binding specificities, thusdemonstrating the utility of the concept.

The CDR-H3 library was used to select antibodies thatbound to the small molecule fluoroscein. Many unique CDR-H3sequences were obtained, but interestingly, many of theseshared significant homology, suggesting a common mechanismof binding. Impressively, the affinities of the best antibodies(0.1mM) approached the average Kd values of the secondaryresponse of immunized mice for free fluoroscein. In a subse-quent study, the concept was expanded to include differentCDR-H3 lengths, and a collection of six libraries (108 cloneseach) was used to select for antibodies that coordinate metal

Synthetic Antibody Libraries 711

ions (6). Antibodies that selectively chelated specific metalions were obtained; most of the functional antibodies werederived from libraries that encoded for long CDR-H3 loops,and the sequences were enriched for amino acids that chelatemetals in natural proteins.

To further expand the diversity of the synthetic libraries,the next step involved the introduction of diversity into bothCDR-H3 and the third CDR of the light chain (CDR-L3) (7). Apanel of libraries (108 clones each) was constructed with diver-sity designs of three basic types: randomization of CDR-H3only with lengths of 5, 10 or 16 amino acids, randomization ofCDR-L3 only with lengths of 8, or 10 amino acids, or simulta-neous randomization of both CDR loops with all possible lengthcombinations of the individual heavy and light chain libraries.This strategy allowed for the direct comparison of the differentdiversity designs against a set of three small molecule haptens.Selections against all three haptens were successful and Fabswith affinities in the nanomolar range were obtained. It wasfound that neither the libraries with only light chain diversitynor those with the shortest CDR-H3 sequences were successfulin generating functional antibodies. Instead, all functional anti-bodies were derived from libraries with the longer CDR-H3lengths and most also contained diversity in CDR-L3. Theseresults suggested that functional antibodies can be obtainedmore readily from relatively incomplete libraries that samplelarge regions of sequence space (i.e., simultaneousdiversificationof multiple, long CDR loops) rather than from more completelibraries that sample a small region of sequence space (e.g.diversification of a single, short CDR loop). The same librarieswere also used to obtain antibodies that recognized DNA withhigh affinity (8), thus demonstrating the applicability of thesynthetic antibody approach to antigens other than small mole-cule haptens.

Another interesting application of synthetic antibodies isthe potential for generating targeted libraries that take intoaccount known binding properties of a particular antigen. Thisapproach was demonstrated at the Scripps Research Instituteby developing antibodies against integrin adhesion recep-tors that recognize natural ligands predominantly through


interactions with a core tripeptide motif (Arg-Gly-Asp, RGD).A library was constructed in which the RGD motif wasembedded within CDR-H3 and was flanked on either side bythree random codons and cysteine residues to promote the for-mation of a disulfide-constrained loop (9). The library wasselected for binding to integrin avb3 and several unique bind-ing clones were obtained. All purified Fabs competed with thenatural ligand vitronectin and exhibited sub-nanomolar bind-ing affinities (Table 1). The Fabs also bound with high affinityto the integrin aIIbb3 but did not recognize the integrin avb5,thus demonstrating some degree of specificity even amongstthe closely related integrin family members. It was noted thatsmall RGD-containing peptides recognize both avb3 and avb5,and thus the specificity observed in the antibodies shows thatappropriate display of the RGD motif within the CDR loopgives rise to receptor specificity. Subsequent work focused onengineering one of the selected Fabs (Fab-9) to discriminatebetween avb3 and aIIbb3 (10). In this case, the RGD motif itselfwas subjected to randomization, and somewhat surprisingly,it was found that the motif is not absolutely required for

Table 1 Affinities of Anti-Integrin Fabs

Affinity (nM)b

Fab CDR-H3a avb3 aIIBb3 avb5

4c CTQGRGDWRSC 0.25 0.25 5007c CTYGRGDTRNC 0.20 0.50 NI8c CPIPRGDWREC 0.20 0.35 NI10c CTWGRGDERNC 0.25 0.25 NI9c CSFGRGDIRNC 1.6 5.0 NIMTF-2d CSFGRTDQRIC 130 17MTF-10d CSFGKGDNRIC 500 7.0MTF-32d CSFGRRDERNC 6.7 2.9MTF-40d CSFGRNDSRNC 11 1.1

aResidues that were randomized in the library are shown in bold text.bAffinities were determined as IC50 values for blocking vitronectin binding (first fourFabs) or as Kd values by surface plasmon resonance.

cFrom Barbas et al. (9).dDerived from Fab-9 by Smith et al. (10).


integrin binding. Antibodies that showed moderate specificityfor aIIbb3 over avb3 were obtained, but antibodies selective foravb3 could not be obtained (Table 1). These studies showedthat targeted synthetic libraries may be used to derive antibo-dies that exhibit affinities and specificities beyond those ofnatural ligands. Antibodies that selectively recognize parti-cular integrins could have major applications as therapeutics,since various integrins have been implicated in osteoporosis,atherosclerosis, metastasis and other pathological states (10).

III. THE MEDICAL RESEARCH COUNCIL

Much pioneeringwork in antibody research has been performedat the Medical Research Council (MRC) at Cambridge in theUnited Kingdom. This work included the development of hybri-doma technology (11) and key contributions to the developmentof methods for the humanization of murine antibodies (12).Winter and colleagues were amongst the first to display anti-body fragments on phage (13,14), and this work lead to thedevelopment of naıve, phage-displayed libraries from naturalrepertoires (Chapter 16), and also, to research into the combina-tion of natural repertoires and synthetic diversity.

The first synthetic antibody library created at the MRCinvolved the introduction of diversity into only CDR-H3 withcompletely random degenerate codons that encoded for all 20natural amino acids (15). Two different libraries were created;one library contained a completely randomized five-residueCDR-H3 while the other contained an eight-residue CDR-H3,in which the first five residues were randomized and the lastthree were fixed as a sequence that commonly occurs in naturalantibodies. For each library, the CDR-H3 diversity was com-bined with 49 germline VH segments that encode most ofthe VH repertoire and a single germline Vl light chain, andrepertoires of modest size (107 clones) were displayed on phagein a single-chain variable fragment (scFv) format. The librarieswere used successfully to isolate antibodies that bound totwo small molecule haptens with modest affinities (Table 2).However, the repertoires were less successful in experiments


with protein antigens, as only a single antibody was obtainedagainst human tumor necrosis factor and no antibodies wereobtained against two other proteins. While these results demo-nstrated the feasibility of combining synthetic diversity withlimited natural diversity, the libraries were not as robust asthose derived from repertoires of V genes rearranged in vivo(16,17). It was concluded that, to obtain antibodies againstany antigen of interest, it was necessary to increase the diver-sity of antigen binding site shapes by increasing the number oflight chains and the diversity of CDR-H3 lengths (15).

The lessons learned from this initial attempt were incor-porated into the design of an improved synthetic library (18).The new library again utilized a single light chain and 50human VH segments, but diversity was significantly expandedby incorporating synthetic CDR-H3 sequences of all lengthsbetween four to 12 residues and by increasing the overalllibrary size by an order of magnitude (108 clones). The librarywas screened against a panel of 18 antigens (includinghaptens and both human and foreign proteins) and bindingclones were obtained in all cases. Sequence analysis of anti-bodies against the range of antigens revealed that most of theVH segments were used to some extent, as were all of theCDR-H3 loop lengths represented in the library. The affinities

Table 2 Affinities of Antibody Fragments Isolated at the MedicalResearch Council

Antigen Kd (nM)a

3-iodo-4-hydroxy-5-nitrophenylacetate (NIP) 700b,d

2-phenyl-5-oxazolone (phOx) 3000b,d

3-iodo-4-hydroxy-5-nitrophenylacetate (NIP) 4.0c,d

Fluorescein 3.8c,d

Fv of monoclonal antibody NQ11 34c,e

Fc of monoclonal antibody NQ11 58c,e

Hepatocyte growth factor=scatter factor 7c,e

aThe best affinity obtained for each antigen is shown.bFrom Hoogenboom and Winter (15).cFrom Griffiths et al. (19).dDetermined by fluorescence quench titration.eDetermined by surface plasmon resonance.


of the antibodies were not measured, but instead, both phageand purified scFvs were used as Western blotting reagents. Ingeneral, the scFv reagents were found to be as sensitive andspecific as monoclonal antibodies. However, the effectivenessof phage-derived scFvs for immunological detection relied onself-aggregation, and it was surmised that multimeric avidityeffects were required to compensate for the modest affinitiesof monomeric antibody fragments obtained from primaryphage repertoires (18).

Further improvements were achieved by dramaticallyincreasing the size of the naıve repertoire using a process ofcombinatorial infection (described in Chapter 3) and display-ing the repertoire in a Fab format (19). Essentially, the heavychain repertoire of Nissim et al. (18) was combined with adiverse light chain repertoire, which consisted of 47 lightchain segments with synthetic CDR-L3 sequences of variablelength containing up to five randomized residues. A bacterialhost harboring the ‘‘donor’’ heavy chain repertoire was infectedwith the ‘‘acceptor’’ light chain repertoire, and the two chainswere combined by Cre catalyzed recombination at loxP sites(20). It was estimated that the recombined repertoire containedclose to 6.5� 1010 different antibodies, and importantly, thediversity was expanded both in terms of absolute numbersand by the addition of light chain diversity.

The recombined library was highly successful in genera-ting specific antibodies against a broad range of haptens andproteins. A total of 215 clones were sequenced, and of these,137 were unique. A detailed analysis of the unique sequencesrevealed some interesting trends in the usage of V gene seg-ments. While a range of V gene segments was used, therewas a clear bias in favor of particular segments in both theheavy and light chains, as only 17 of 49 heavy chain segmentsand 19 of 47 light chain segments were used by the functionalbinding clones. In particular, almost half of the VH genes usedVH segment DP-47, which is also most common in naturalantibodies. A single Vk segment (DPK-15) and a single Vl seg-ment (DPL-3) accounted for the majority of the light chains,but in contrast with the heavy chain, these segments arenot common in natural antibodies. Overall, these results


suggested that particular V genes (e.g. DP-47) are better ableto generate functional antigen-binding sites that can recog-nize a diverse range of antigens. Thus, while it may be advan-tageous to include multiple V genes in synthetic libraries, it islikely that a small subset of the natural V genes can accom-modate most, if not all, of the diversity required of a robustantibody library.

The binding affinities were determined for several Fabproteins raised against two small molecule haptens, fluoures-cein and 3-iodo-4-hydroxy-5-nitrophenyl-acetate (NIP). Thebest affinities for soluble hapten were found to be in the lownanomolar range (Table 2). Binding was also assayed for alimited number of Fabs that bound to two protein antigens(monoclonal antibodyNQ11=7.22andhepatocytegrowth factor=scatter factor), and again, the best affinities were found to bein the low nanomolar range (Table 2). These results demon-strated that large synthetic repertoires may be used directlyto produce high affinity antibodies against both haptens andlarge proteins.

The concepts explored byWinter and colleagues have alsobeen extended by other groups. Logtenberg and colleaguesconstructed a library that contained the same 49 germ-line VH genes described above and seven light chains (21).Synthetic diversity was introduced into CDR-H3 by insertingrandomized loops ranging from 6–15 residues in length. Thecentral regions of these loops were completely randomized,but variability at the borders was restricted to only thosesequences that are commonly observed in natural antibodies,as it was reasoned that the functional diversity of the librarywould be improved by favoring sequences that are abundantin natural repertoires. Specific binding clones were isolatedfor 13 different antigens but, perhaps because the librarywas only of moderate size (3.6� 108 clones), binding affinitieswere rather modest (2mM to 100nM).

Neri and colleagues have used principles of proteindesign to further focus diversity at regions most likely tobe involved in functional interactions with antigen (22). Alibrary was constructed using a single VH (DP-47) and asingle Vk (DP-K22) that dominate the natural repertoire


(23). Four positions each in CDR-H3 (positions 95–98) andCDR-L3 (positions 91, 93, 94 and 96) were chosen as sitesfor the introduction of diversity (Fig. 1), as residues at thesepositions commonly make contacts with antigen (3). The eightsites were completely randomized to produce a naıve libraryof modest size (3� 108 clones), and specific binding cloneswere isolated against a panel of six protein antigens. Anti-bodies isolated against the ED-B domain of fibronectin wereanalyzed in detail and affinities were found to be in thedouble-digit nanomolar range. To demonstrate the advan-tages of the simplified library design, one of these antibodieswas subjected to affinity maturation by randomizing heavychain sites located on the periphery of the antigen-bindingsite (Fig. 1). The new library yielded an antibody with 27-foldimproved affinity (Kd¼ 1.5nM). This improved antibody wasused for a further round of affinity maturation in which two

Figure 1 The library design of Pini et al. (22) The main-chains ofan antibody variable domain are shown as grey ribbons. The Caatoms of CDR residues included in the libraries are shown asspheres colored black (naıve library) or grey (affinity maturationlibraries). Numbering follows the Kabat nomenclature (47). Heavyor light chain residues are numbered in bold text or italics, respec-tively. Structures were drawn with PyMOL (DeLano Scientific, SanCarlos, CA).


sites in the light chain were targeted for randomization,and ultimately, an ultra-high affinity antibody was obtained(Kd¼ 54 pM). In an interesting extension of this approach, asimilar library was constructed using a scFv framework thatis stable in the reducing environment of the cytoplasm (24).The library yielded specific antibodies against a variety ofantigens, and importantly, the thermodynamic stabilities ofthese antibodies under reducing conditions were similar tothat of the parent antibody. Taken together, these studiesdemonstrate the power of synthetic antibody libraries con-structed on the basis of rational design principles for boththe choice of scaffold and for sites of diversification.

A further extension of the utility of synthetic antibodylibraries has been to engineer simultaneously both antigen-binding activity and the global structure of the antibodyfragment. While the antigen-binding sites of most naturalantibodies consist of a heterodimer of heavy and light chains,

Figure 2 Crystal structure of the autonomous VH domain HEL4(PDB entry 1OHQ). The hydrophobicity of the former light chaininterface is reduced significantly by burial of the Trp47 side-chaininto a cavity formed by Gly35, Val37 and Ala93. The main-chainCa trace is shown as a ribbon and selected side-chains are shownas sticks. Residues are numbered according to the Kabat nomencla-ture (47), and labeled residues are colored black.


certain camelid antibodies are devoid of the light chain, andinstead, contain VH domains that fold and function auto-nomously (25–28). It has been shown that the autonomousnature of camelid VH domains is, at least in part, due to substi-tutions in the region of the VH domain that formerly madecontact with the light chain (29). These substitutions arebelieved to increase the hydrophilicity of the former light chaininterface and thus reduce aggregation. Attempts to produceautonomous human VH domains by incorporating camelidsubstitutions into the framework have been moderately suc-cessful, but the substitutions tend to cause thermodynamicdestabilization and do not completely eliminate aggregation(30–33). In addition, the introduction of camelid substitutionsinto human antibodies increases the risk of immunogenicityin vivo. In order to produce autonomous human VH domains,Winter and colleagues have applied a synthetic library strat-egy (34). A library was generated by randomizing residues inall three heavy chains CDRs of a human VH domain that wasdisplayed on phage without a light chain. The large library(1.1� 1010 clones) was screened for autonomous VH domainsthat bound to hen egg white lysozyme. Crystallographic analy-sis of one such autonomous VH domain (HEL4) revealed thatTrp47 at the former light chain interface tucks into a hydro-phobic cavity formed by Gly35, Val37 and Ala93 (Fig. 2), andthis significantly increases the hydrophilicity of the exposedinterface. Gly35 resides at a position within CDR-H1 thatwas randomized in the library, and the lack of a side-chain atthis site was critical in forming a cavity for Trp47. Thus, a func-tional, autonomous human VH domain was derived withoutresorting to camelid framework substitutions, and theseresults demonstrate how synthetic libraries can be used toengineer not only novel binding specificities, but also, to alterbasic aspects of the immunoglobulin fold.

IV. MORPHOSYS

Researchat thebiotechnologycompanyMorphosys (Martinsried,Germany) stems from the work of Pluckthun and colleagues


at the University of Zurich, and has focused on developingsynthetic antibody libraries based on a structural analysisof the human antibody repertoire (35). Antibody frameworkswere chosen on the basis of structural stability and highexpression, and the genes were synthesized chemically. Highexpression was ensured by the use of common Escherichiacoli codons and subsequent affinity maturation was facili-tated by the strategic placement of unique restriction sites.Sequence and structural analysis of natural antibody data-bases was used to identify a limited number of frameworksthat would efficiently cover the full range of CDR diversityin the human immune repertoire. In total, seven genes eachwere chosen to represent heavy and light chain diversity, andthese provided 49 combinations of light and heavy chainsthat were displayed on phage in a scFv format.

The first Morphosys human combinatorial antibodylibrary (termed ‘‘HuCAL1’’) was constructed by randomizingboth CDR-H3 and CDR-L3 at the center of the antigen-binding site (35). To maximize the probability of obtainingfunctional antibodies with CDR sequences resembling thoseof natural antibodies, sequences from human antibodies wereanalyzed to obtain an amino acid distribution at each positionto be randomized. In the case of the light chain, Vk and Vl

sequences were analyzed separately, while all VH sequenceswere grouped together (Fig. 3). Oligonucleotides were thensynthesized to mimic the natural amino acid distribution,using a trinucleotide synthesis method that allows for precisetailoring of diversity at each position (36). For the Vk CDR-3,a single length that occurs most often in nature was used;three positions that point into the antigen-binding site werehighly randomized and four others were mildly randomized.For the Vl CDR-3, three different lengths were used; oneposition was mildly randomized and four, five or six positionswere highly randomized. As the VH CDR-3 varies greatlyin length and sequence, the library was designed to containall possible loop lengths between five and 28 residues and,except for positions near the C-terminal base of the loop thatwere mildly randomized, all positions were highly rando-mized to contain the overall amino acid composition of natural


CDR-H3 sequences. TheHuCAL1was a relatively large library(2.1� 109 clones), and DNA sequencing of the naıve reper-toire revealed that diversity at each of the designed CDRsites closely mimicked that of the natural immune repertoire.

In the original report, the HuCAL1 was screened againsta variety of antigens, but results were given for only a fewexamples (35). Affinities for peptide antigens were found tobe in the micromolar range, whereas affinities for proteinantigens were in the nanomolar range (Table 3). Extremelyhigh affinities in the picomolar range were achieved against

Figure 3 Composition of the CDR-3 sequences of HuCAL1. Thefrequency of each of the twenty natural amino acids at each sitethat was randomized in the heavy chains (VH) and the light chains(Vk and Vl) was determined by sequencing 257 clones. Numberingfollows the Kabat nomenclature (47); for CDR-H3 the position pre-ceding position 101 is numbered 100z and the length variable regionis numbered 95–100s.


bovine insulin, but the selections involved a ribosome displayprocess during which point mutations accumulated andimproved affinity up to 40-fold compared to the naıve progeni-tor (35). Subsequent reports have focused on exploiting themodular design of the library to facilitate the high throughputproduction of antibodies (37,38). Automated panning andscreening systems that facilitated the analysis of multipleantigens were developed, and in addition, library screeningwas performed directly on mammalian cells expressing sur-face antigens. The modularity of the library was exploited toreformat the scFvs into different forms, including Fabs andfull-length immunoglobulins. Overall, the library was foundto be extremely robust, as the success rate was high, providedthe antigen was of high quality. The isolated antibodies wereused for a variety of applications, including flow cytometry,immunoprecipitation, western blotting and immunohisto-chemistry; affinities were determined for scFvs against three

Table 3 Affinities of Antibody Fragments Isolated from theMorphosys HuCAL

Antigen Kd (nM)a

Intracellular adhesion molecule-1 9.4b

CD11b 1.0b

Epidermal growth factor receptor 246b

Mac1 peptide 1130b

Hag peptide 610b

NFB peptide 1600b

Bovine insulin 0.082c

Integrin Mac-1 1.7d

Fibroblast growth factor receptor-3 0.7e

Tissue inhibitor of metalloproteinase-1 13f

Citrate transporter CitS 4g

aAffinities were determined by surface plasmon resonance, and the best affinity foreach antigen is shown.

bFrom Knappik et al. (35).cFrom Hanes et al. (45), obtained by ribosome display and random mutagenesis.dFrom Krebs et al. (37).eFrom Rauchenberger et al. (40).fFrom Parsons et al. (41).gFrom Rothlisberger et al. (42).


protein antigens and were found to be in the low to high nano-molar range (Table 3).

The major impetus for the HuCAL approach was to pro-duce synthetic antibody libraries that closely resembled thehuman repertoire, so potential therapeutic antibodies couldbe obtained rapidly (35). Several reports have describedefforts in this direction. The HuCAL1was used to obtain scFvsthat recognize human leukocyte antigen-DR (HLA-DR), as afirst step in developing a therapeutic that induces pro-grammed death of lymphoma=leukemia cells expressing theantigen (39). Twelve unique clones were obtained, and oneof these was subjected to an affinity maturation process thatdemonstrated the power of the modular HuCAL design,which allowed for the rapid generation of secondary librariesby using the strategically placed restriction sites to introduceadditional CDR diversity. Clone scFv-B8 was converted to aFab format, CDR-L3 was re-randomized, and the rather mod-est affinity of the parental clone (Kd¼ 350nM) was improvedby approximately 6-fold. In a second round of optimization,additional diversity was introduced into CDR-L1, severalclones with affinities in the low nanomolar range were iso-lated, and these were converted into IgG4 antibodies withsub-nanomolar affinities (Table 4). The antibodies exhibitedpotent tumoricidal activity, both in vitro in assays with lym-phoma and leukemia cell lines, and in vivo in xenograft mod-els of non-Hodgkin lymphoma. A Fab version of HuCAL1(HuCAL-Fab1) has also been constructed and was used toobtain potent neutralizing antibodies against human fibro-blast growth factor receptor-3 (Table 3), a potential therapeu-tic antibody target that had previously proven recalcitrant tothe generation of blocking antibodies (40). The HuCAL-Fab1has also been used to obtain neutralizing antibodies againsttissue inhibitor of metalloproteinase-1 (Table 3), and theseantibodies have been shown to be efficacious in attenuatingliver fibrosis in an animal model, by relieving the inhibitionof metalloproteinases that are involved in the degradationof extracellular matrix within injured liver tissue (41).

Aside from the development of therapeutic leads,Pluckthun and colleagues have also applied theHuCAL concept


to novel applications. The determination of crystal structuresfor membrane proteins remains a daunting challenge forstructural biology, and it is believed that the use of antibodyfragments as crystallization chaperones may greatly aid theseefforts. As a proof of concept, an Fab library in which all sixCDRs were randomized according to the HuCAL strategywas used to derive Fabs against the detergent-solubilizedcitrate transporter CitS of Klebsiella pneumoniae (42). Toensure that the Fabs would be highly expressed and stablein detergent, the library was restricted to a subset of theHuCAL that contained only the most stable combination oflight and heavy chain frameworks (43,44). An Fab with lownanomolar affinity for CitS was obtained (Table 3), andalthough crystallization was not reported, it was shown thatthe Fab:CitS complex could be detected and purified by gelfiltration chromatography. The HuCAL1 was also used toisolate catalytic antibodies with phosphatase activity, usinga turnover-based in vitro selection that relied on a substratetrap for covalent capture of the catalytic antibody (45). Thenaıve repertoire yielded a scFv that exhibited a catalytic pro-ficiency that was approximately 100-fold greater than that ofthe best phosphatase-like antibody obtained by hybridomamethods (46), and activity was improved a further ten-foldby random mutagenesis and directed evolution of the par-ental clone. These results demonstrated that large synthetic

Table 4 Affinity Optimization of Anti-HLA-DR Antibodies

Antibody Format Optimizationa Kd (nM)b

B8 Fab parent 3507BA Fab L3 60305D3 Fab L3þL1 3.01C7277 Fab L3þL1 3.01D09C3 Fab L3þL1 3.0305D3 IgG4 L3þL1 0.51C7277 IgG4 L3þL1 0.61D09C3 IgG4 L3þL1 0.3

aIndicates additional light chain CDRs that were randomized.bDetermined by surface plasmon resonance.


repertoires and controlled in vitro selections can yield cataly-tic antibodies with activities far greater than those obtainedfrom natural repertoires by immunization.

V. GENENTECH

Synthetic antibody libraries atGenentech (South SanFrancisco,CA) have been developed with a single human frameworkderived from the consensus sequences of the most abundanthuman subclasses, namely VH subgroup III and Vk sub-group I (47). This framework was originally developed forthe humanization of murine antibodies, and in fact, it hasbeen used in several antibodies that are successful thera-peutics (48–51). The anti-ErbB2 antibody humanized 4D5(48) was chosen as the library scaffold, because Fab4D5 iswell expressed in E. coli and had been displayed previouslyon phage (52), and in addition, the high-resolution crystalstructure of the variable fragment has been determined(53).

In the first libraries, the antibody was displayed as amonomeric scFv, because the scFv format usually results inhigher levels of display relative to the Fab format (54). A biva-lent scFv display format was also developed by inserting adimerization domain between the scFv and the phage coatprotein. Bivalent binding produces an avidity effect thatincreases apparent binding affinities for immobilized antigens(55,56), and it was reasoned that this would improve therecovery of rare or low affinity clones from naıve libraries.

To simplify library design and construction, only theheavy chain was subjected to randomization, and structuralanalysis was used to identify sites that would be suitable forsupporting diversity (Fig. 4). Within CDR-H1 and CDR-H2,solvent-exposed sites that were positioned for potential con-tact with antigen were chosen and buried residues wereexcluded to minimize structural perturbations. Within thehypervariable CDR-H3, a continuous stretch of residueswithin the central region of the loop was chosen. For each sitewithin CDR-H1 and CDR-H2, tailored degenerate codons


were designed to bias the diversity in favor of amino acidtypes commonly found in the human repertoire (Fig. 5), asit was reasoned that this would produce ‘‘human-like’’ CDRsequences. As there is little position-specific bias within nat-ural CDR-H3 sequences, the entire loop was randomized witha degenerate codon that encoded for 12 amino acids, butexcluded most aliphatic side-chains that may give rise tonon-specific hydrophobic interactions.

Highly optimized library construction methods (57) wereused to generate very large libraries (5� 1010 clones) in boththe monovalent and bivalent format. The libraries werehighly successful against a panel of six protein antigens, inthat numerous binding clones were obtained in each case.

Figure 4 CDR residues chosen for diversification in the Genen-tech synthetic antibody libraries. The main-chains of the huma-nized 4D5 VH and VL domains are shown as grey tubes. The Caatoms of CDR residues included in the libraries are shown asspheres colored black (VH) or grey (VL). Numbering follows theKabat nomenclature.


In particular, the bivalent format yielded dozens or even hun-dreds of unique clones against each antigen. Analysis of 86unique anti-murine vascular endothelial growth factor(mVEGF) clones revealed a broad range of affinities withthe tightest binders in the 100nM range. Interestingly, thebivalent format allowed for the selection of numerous clones,while the monovalent format selected only the few clones withthe highest affinities. Thus, the two formats were shown to becomplementary; the monovalent format is well suited for

Figure 5 Diversity of CDR-H1 and CDR-H2 in the Genentechsynthetic libraries. For each site chosen for diversification, a tai-lored degenerate codon (italics) was designed to encode amino aciddiversity similar to that found in natural antibodies. Numbering fol-lows the Kabat nomenclature (47). Equimolar DNA degeneraciesare represented in the IUB code (D¼A=G=T, K¼G=T, M¼A=C,N¼A=C=G=T, R¼ A=G, S¼G=C, V¼A=C=G, W¼A=T, Y¼C=T).


stringent affinity selections, while the bivalent format is use-ful for the recovery of rare clones against difficult antigens.

In a follow-up study, many different CDR designs wereexplored in a bivalent Fab format (58). It was found thatlibrary performance was improved significantly by furtherreducing the diversity in CDR-H1 and CDR-H2, and concur-rently increasing the diversity in CDR-H3. Ultimately, ahighly robust library was constructed in which the CDR-H1and CDR-H2 diversity was highly restricted to only thoseamino acids that are most common in natural antibodies,and the CDR-H3 diversity was greatly expanded to includecombinations of all 20 natural amino acids in loop lengths ofseven to 19 residues. This library was screened against apanel of 13 antigens and binding clones were found in allcases; for nine of the antigens, Fabs with affinities better than10nM were obtained (Table 5). As only the heavy chain hadbeen randomized in the naıve library, it was shown that lightchain diversity could be exploited for subsequent affinitymaturation. A high affinity anti-mVEGF heavy chain(Kd¼ 0.9 nM) was combined with a light chain librarydesigned along principles similar to those used for the heavychain design. Many light chain sequences were found toimprove affinity, and the best Fab bound to mVEGF withan affinity in the low picomolar range (Kd¼ 20 pM). Takentogether, these results demonstrated that a single scaffoldlibrary with diversity restricted to the heavy chain can yieldhigh affinity binders to most protein antigens, provided

Table 5 Affinities of Antibody Fragments Isolated at Genentech

Antigena IC50 (nM)b

Vascular endothelial growth factor 0.6NKG2D extracellular domain 3April 1.4Immunoglobulin E 9.8Tissue factor 11

aAll antigens were of murine origin.bThe best affinity for each antigen is shown and was determined by competitiveenzyme-linked immunosorbant assay (ELISA).


that the diversity is tailored to favor functional antibodysequences. Furthermore, the light chain can be recruitedin a second step to obtain ultra-high affinity synthetic anti-bodies.

Single framework libraries have also been used to inves-tigate the basic principles governing antigen recognition.Whereas natural CDR sequences are highly diverse, thereare clear biases for particular amino acids (59–65). Tyrosineand serine are particularly abundant, and tyrosine residuescontribute a disproportionate number of antigen contacts(3,66,67). In an effort to determine the minimal requirementsfor antigen recognition, a heavy chain library was constructedby allowing random combinations of only four amino acid (tyr-osine, aspartate, alanine and serine) (68). Despite the limitedchemical diversity, specific antibodies with affinities in themicromolar range were obtained against human VEGF, anangiogenic hormone that has been implicated in tumorogen-esis (69,70). The selected heavy chains were combined witha light chain library and Fabs with affinities in the low nano-molar range were obtained (Kd¼ 2–10nM). Crystallographicanalysis of two distinct Fabs (YADS1 and YADS2) in complexwith hVEGF revealed that antigen recognition was mediatedprimarily by tyrosine residues which comprised 71% of thestructural epitope, and in contrast, aspartate was almostentirely excluded from the binding sites (Fig. 6).

These results led to further restriction of the chemicaldiversity to a binary code containing only tyrosine and serine(71). A degenerate codon that encodes for tyrosine and serinewas used to randomize solvent-accessible sites in CDR-H1and CDR-H2 and random loops of variable lengths wereinserted in place of CDR-H3. Two large libraries (1010 clones),which differed only in that one contained a randomized CDR-L3,were constructed and screened against a panel of six proteinantigens. Despite the extreme restrictions on chemical diver-sity, specific binding clones were obtained against each anti-gen. Detailed functional analysis of anti-hVEGF antibodiesrevealed that the Fabs bound with surprisingly high affinity(Kd¼ 60nM) and were as specific as natural antibodies in cell-based assays. Structural analysis of a binary Fab (Fab-YSd1)


raised against human death receptor DR5, a cell-surfacereceptor that mediates apoptotic cell death (72), revealedthe structural basis for the minimalist molecular recognition.While the Fab interacts with hDR5 in a normal manner, theunusually long CDR-H3 protrudes from the framework andmakes extensive contact with antigen (Fig. 7A). Interest-ingly, the CDR-H3 loops contains a ‘‘biphasic’’ helix with tyro-sine and serine clustered on opposite faces (Fig. 7B). Thetyrosine face is buried against the surface of hDR5 and con-tributes all of the CDR-H3 accessible surface buried uponantigen binding, and serine residues likely play structural

Figure 6 Anti-hVEGF antibodies obtained from a four-amino-acidcode. (A) The complex of hVEGF with Fab-YADS1 (left panel, PDBID code 1TZH) and Fab-YADS2 (right panel, PDB ID code 1TZI).The antigen is depicted as a white molecular surface. The main-chains of the Fabs are shown as cartoons; the heavy and lightchains are colored black or grey, respectively. (B) The CDR side-chains of Fab-YADS1 (left panel) and Fab-YADS2 (right panel) thatcontact hVEGF. The Fab side-chains are shown as sticks coloredblack (tyrosine) or grey (all others).


roles. Overall, these results demonstrate that functional anti-bodies can be created with even the simplest chemical diver-sity, and in fact, it is advantageous to restrict diversity tosmall subsets of functional groups that are particularly wellsuited for mediating molecular recognition.

VI. CONCLUSIONS

While progress in synthetic antibody research has beenrelatively slow in comparison with that of natural phage-displayed repertoires, significant advances have been made.In fact, libraries developed over the last few years haveachieved performance levels equivalent to those of theirnatural counterparts. As further insights are gained intothe basic principles of library design and function, furtherprogress can be expected. Thus, it seems likely that universalsynthetic antibody libraries that can provide binding speci-ficities against any antigen of interest will be available in thenear future. These libraries hold great promise for the

Figure 7 An anti-hDR5 antibody obtained from a binary code. (A)The complex of hDR5 with Fab-YSd1 (PDB 1ZA3). The antigen isdepicted as a white molecular surface. The main-chains of the Fabare shown as cartoons; the heavy and light chains are colored blackor grey, respectively. The CDR-H3 loop is circled. (B) Contactsbetween CDR-H3 and hDR5. The CDR-H3 main-chain is depictedas a cartoon and side-chains are rendered as sticks.


development ofhighlyoptimizedreagentsand therapeutics, andin addition, will be extremely useful tools for elucidating thebasic principles that govern antibody structure and function.

REFERENCES

1. Barbas CF III, Bain JD, Hoekstra DM, Lerner RA. Semisyn-thetic combinatorial antibody libraries: a chemical solution tothe diversity problem. Proc Natl Acad Sci USA 1992;89:4457–4461.

2. Xu JL, Davis MM. Diversity in the CDR3 region of V(H) issufficient for most antibody specificities. Immunity 2000; 13:37–45.

3. Padlan EA. Anatomy of the antibody molecule. Mol Immunol1994; 31:169–217.

4. Wilson IA, Stanfield RL. Antibody-antigen interactions: newstructures and new conformational changes. Curr Opin StructBiol 1994; 4:857–867.

5. Chothia C, LeskAM. Canonical structures for the hyper variableregions of immunoglobulins. J Mol Biol 1987; 196:901–917.

6. Barbas CF III, Rosenblum JS, Lerner RA. Direct selection ofantibodies that coordinate metals from semisynthetic combina-torial libraries. Proc Natl Acad Sci USA 1993; 90:6385–6389.

7. Barbas CF III, Amberg W, Simoncsits A, Jones TM, LernerRA. Selection of human anti-hapten antibodies from semisyn-thetic libraries. Gene 1993; 137:57–62.

8. Barbas SM, Ghazal P, Barbas CF III, Burton DR. Recognitionof DNA by synthetic antibodies. J Am Chem Soc 1994;116:2161–2162.

9. Barbas CF III, Languino LR, Smith JW. High-affinity self-reactive human antibodies by design and selection: targetingthe integrin ligand binding site. Proc Natl Acad Sci USA1993; 90:10003–10007.

10. Smith JW, Hu D, Satterthwait A, Pinz-Sweeney S, Barbas CFIII,: Building synthetic antibodies as adhesive ligands forintegrins. J Biol Chem 1994; 269:32788–32795.


11. Kohler G, Milstein C. Continuous cultures of fused cells secret-ingantibody of predefinedspecificity.Nature1975; 256:495–497.

12. Jones PT, Dear PH, Foote J, Neuberger MS, Winter G. Repla-cing the complementarity-determining regions in a humanantibody with those from a mouse. Nature 1986; 321:522–525.

13. Hoogenboom HR, Griffiths AD, Johnson KS, Chiswell DJ,Hudson P, Winter G. Multi-subunit proteins on the surface offilamentous phage: methodologies for displaying antibody (Fab)heavy and light chains. Nucl Acids Res 1991; 19:4133–4137.


15. HoogenboomHR,Winter G. By-passing immunization: Humanantibodies from synthetic repertoires of germline VH genesegments rearranged in vitro. J Mol Biol 1992; 227:381–388.

16. Marks JD, Hoogenboom HR, Bonnert TP, McCafferty J,Griffiths AD, Winter G. By-passing immunization: humanantibodies from V-gene libraries displayed on phage. J MolBiol 1991; 222:581–597.

17. Griffiths AD, Malmqvist M, Marks JD, Bye JM, Embleton MJ,McCafferty J, Baier M, Holliger KP, Gorick BD, Hughes-JonesNC, et al. Human anti-self antibodies with high specificityfrom phage display libaries. EMBO J 1993; 12:725–734.

18. Nissim A, HoogenboomHR, Tomlinson IM, Flynn G,Midgley C,Lane D, Winter G. Antibody fragments from a ‘single pot’ phagedisplay library as immunochemical reagents. EMBO J 1994;13:692–698.

19. Griffiths AD, Williams SC, Hartley O, Tomlinson IM,Waterhouse P, Crosby WL, Kontermann RE, Jones PT, LowNM, Allison TJ, Prospero TD, Hoogenboom HR, Nissim A,Cox JPL, Harrison JL, Zaccolo M, Gheradi E, Winter G. Isola-tion of high affinity human antibodies directly from largesynthetic repertoires. EMBO J 1994; 13:3245–3260.

20. Waterhouse P, Griffiths AD, Johnson KS, Winter G. Combina-torial infection and in vivo recombination: a strategy formaking large phage antibody repertoires. Nucleic Acids Res1993; 21:2265–2266.


21. de Kruif J, Boel E, Logtenberg T. Selection and applicationof human single chain Fv antibody fragments from a semi-synthetic phage antibody display library with designed CDR3regions. J Mol Biol 1995; 248:97–105.

22. Pini A, Viti F, Santucci A, Carnemolla B, Zardi L, Neri P,Neri D. Design and use of a phage display library: human anti-bodies with subnanomolar affinity against a marker of angio-genesis eluted from a two-dimensional gel. J Biol Chem1998; 273:21769–21776.

23. Kirkham PM, Mortari F, Newton JA, Schroeder HWJ. Immu-noglobulin VH clan and family identity predicts variable domainstructure and may influence antigen binding. EMBO J 1992; 11:603–609.

24. DesiderioA,FranconiR,LopezM,VillaniME,VitiF,ChiaraluceR,Consalvi V, Neri D, Benvenuto E. A semi-synthetic repertoire ofintrinsically stable antibody fragments derived from a single-framework scaffold. J Mol Biol 2001; 310:603–615.

25. Hamers-CastermanC,AtarhouchT,MuyldermansS, RobinsonG,Hamers C, Songa EB, Bendahman N, Hamers R. Naturallyoccurring antibodies devoid of light chains. Nature 1993; 363:446–448.

26. Ewert S, Cambillau C, Conrath K, Pluckthun A. Biophysicalproperties of camelid V(HH) domains compared to those ofhuman V(H)3 domains. Biochemistry 2002; 41:3628–3636.

27. Dumoulin M, Conrath K, Van Meirhaeghe A, Meersman F,Heremans K, Frenken LG, Muyldermans S, Wyns L,Matagne A. Single-domain antibody fragments with highconformational stability. Nature Struct Biol 2002; 11:500–515.

28. Arbabi Ghahroudi M, Desmyter A, Wyns L, Hamers R,Muyldermans S. Selection and identification of single domainantibody fragments from camel heavy-chain antibodies. FEBSLett 1997; 414:521–526.

29. Muyldermans S, Atarhouch T, Saldanha J, Barbosa JARG,Hamers R. Sequence and structure of VH domain from natu-rally occurring camel heavy chain immunoglobulins lackinglight chains. Protein Eng 1994; 7:1129–1135.


30. Martin F, Volpari C, Steinkuhler C, Dimasi N, Brunetti M,Biasiol G, Altamura S, Cortese R, De Francesco R, Sollazzo M.Affinity selection of a camelized V(H) domain antibody inhibitorof hepatitis C virus NS3 protease. Protein Eng 1997; 10:607–614.

31. Riechmann L. Rearrangement of the former VL interface inthe solution structure of a camelised, single antibody VHdomain. J Mol Biol 1996; 259:957–969.

32. Davies J, Riechmann L. ‘Camelising’ human antibody frag-ments: NMR studies on VH domains. FEBS Lett 1994; 339:285–290.

33. Davies J, Riechmann L. Single antibody domains as smallrecognition units: design and in vitro antigen selection ofcamelized, human VH domains with improved protein stabi-lity. Protein Eng 1996; 9:531–537.

34. Jespers L, Schon O, James LC, Veprintsev D, Winter G. Crys-tal structure of HEL4, a soluble refoldable human VH singledomain with a germ-line scaffold. J Mol Biol 2004; 337:893–903.

35. Knappik A, Ge L, Honegger A, Pack P, FischerM,Wellnhofer G,Hoess A, Wolle J, Pluckthun A, Virnekas B. Fully synthetichuman combinatorial antibody libraries (HuCAL) based onmodular consensus frameworks and CDRs randomized withtrinucleotides. J Mol Biol 2000; 296:57–86.

36. Virnekas B, Ge L, Pluckthun A, Schneider KC, Wellnhofer G,Moroney SE. Trinucleotide phosphoramidites: ideal reagentsfor the synthesis of mixed oligonucleotides for randommutagenesis. Nucl Acids Res 1994; 22:5600–5607.

37. Krebs B, Rauchenberger R, Reifert S, Rothe C, Tesar M,Thomassen E, Cao M, Dreier T, Fischer D, Hob A, Inge L,Knappik A, Margit M, Pack P, Meng X-Q, Schier R, Sohle-mann P, Winter J, Wolle J, Kretschmar T. High-throughputgeneration and engineering of recombinant human antibodies.J Immunol Methods 2001; 254:67–84.

38. Frisch C, Brocks B, Ostendorp R, Hoess A, von Ruden T,Kretzschmar T. From EST to IHC: human antibody pipelinefor target research. J Immunol Methods 2003; 75:203–212.


39. Nagy ZA, Hubner B, Lohning C, Rauchenberger R, Reiffert S,Thomassen-Wolf E, Zahn S, Leyer S, Schier EM, Zahradnik A,Brunnerr C, Lobenwein K, Rattel B, Stanglmaier M, Hallek M,Wing M, Anderson S, Dunn M, Kretschmar T, Tesar M. Fullyhuman, HLA-DR-specific monoclonal antibodies efficientlyinduce programmed death of malignant lymphoid cells. Nat-ure Medicine 2002; 8:801–807.

40. Rauchenberger R, Borges E, Thomassen-Wolf E, Rom E,Adar R, Yaniv Y, Malka M, Chumakov I, Kotzer S, ResnitzkyD, Knappik A, Reiffert S, Prassler J, Jury K, Waldherr D,Bauer S, Kretschmar T, Yayon A, Rothe C. Human combina-torial Fab library yielding specific and functional antibodiesagainst the human fibroblast growth factor receptor 3. J BiolChem 2003; 278:38194–38205.

41. Parsons CJ, Bradford BU, Pan CQ, Cheung E, Schauer M,Knorr A, Krebs B, Kraft S, Zahn S, Brocks B, Feirt B, Mei B,Cho M-S, Ramamoorthi R, Roldan G, Ng P, Lum P, Hirth-Dietrich C, Tomkinson A, Brenner DA. Antifibrotic effects of atissue inhibitor of metalloproteinase-1 antibody on establishedliver fibrosis in rats. Hepatology 2004; 40:1106–1115.

42. Rothlisberger D, Pos KM, Pluckthun A. An antibody libraryfor stabilizing and crystallizing membrane proteins - selectingbinders to the citrate carrier CitS. FEBS Lett 2004; 564:340–348.

43. Ewert S, Huber T, Honegger A, Pluckthun A. Biophysicalproperties of human antibody variable domains. J Mol Biol2003; 325:531–553.

44. Ewert S, Honegger A, Pluckthun A. Stability improvement ofantibodies for extracellular and intracellular applications:CDR grafting to stable frameworks and structure-basedframework engineering. Methods 2004; 34:148–199.

45. Hanes J, Schaffitzel C, Knappik A, Pluckthun A. Picomolaraffinity antibodies from a fully synthetic naive library selectedand evolved by ribosome display. Nature Biotechnol 2000;18:1287–1292.

46. Scanlan TS, Prudent JR, Schultz PG. Antibody-catalyzedhydrolysis of phosphate monoesters. J Am Chem Soc 1991;113:9397–9398.


47. Kabat EA, Wu TT, Perry HM, Gottesman KS, Foeller C.Sequences of Proteins of Immunological Interest 5th ed.Bethesda, MD: National Institutes of Health 1991.

48. Carter P, Presta L, Gorman CM, Ridgway JB, Henner D,Wong WL, Rowland AM, Kotts C, Carver ME, Shepard HM.Humanization of an anti-p185HER2 antibody for human can-cer therapy. Proc Natl Acad Sci USA 1992; 1992:4285–4289.

49. Presta LG, Lahr SJ, Shields RL, Porter JP, Gorman CM,Fendly BM, Jardieu PM. Humanization of an antibody direc-ted against IgE. J Immunol 1993; 151:2623–2632.

50. Presta LG, Chen H, O’Connor SJ, Chisholm V, Meng YG,Krummen L, Winkler M, Ferrara N. Humanization of ananti-vascular endothelial growth factor monoclonal antibodyfor the therapy of solid tumors and other disorders. CancerRes 1997; 57:4593–4599.

51. Werther WA, Gonzalez TN, O’Connor SJ, McCabe S, Chan B,Hotaling T, ChampeM, Fox JA, Jardieu PM, BermanPW, PrestaLG. Humanization of an anti-lymphocyte function-associatedantigen (LFA)-1 monoclonal antibody and reengineering of thehumanized antibody for binding to rhesus LFA-1. J Immunol,1996, 157:4986–4995.

52. Garrard LJ, Henner DJ. Selection of an anti-IGF-1 Fab from aFab phage library created by mutagenesis of multiple CDRloops. Gene 1993; 128:103–109.

53. Eigenbrot C, Randal M, Presta L, Carter P, Kossiakoff AA.X-ray structures of the antigen-binding domains from threevariants of humanized anti-p185HER2 antibody 4D5 andcomparison with molecular modeling. J Mol Biol 1993; 229:969–995.

54. Sidhu SS, Li B, ChenY, Fellouse FA, Eigenbrot C, FuhG. Phage-displayed antibody libraries of synthetic heavy chain comple-mentarity determining regions. J Mol Biol 2004; 338:299–310.

55. Pack P, Pluckthun A. Miniantibodies: use of amphipathichelices to produce functional, flexibly linked dimeric Fvfragments with high avidity in Escherichia coli. Biochemistry1992; 31:1579–1584.


56. Lee CV, Sidhu SS, Fuh G. Bivalent antibody phage displaymimics natural immunoglobulins. J Immunol Methods 2004;284:119–132.

57. Sidhu SS, Lowman HB, Cunningham BC, Wells JA. Phagedisplay for selection of novel binding peptides.Methods Enzymol2000; 328:333–363.

58. Lee CV, LiangW-C, Dennis MS, Eigenbrot C, Sidhu SS, Fuh G.High-affinity human antibodies from phage-displayed syn-thetic Fab libraries with a single framework scaffold. JMol Biol2004; 340:1073–1093.

59. Lea S, Stuart D. Analysis of antigenic surfaces of proteins.FASEB J 1995; 9:87–93.

60. Kabat EA, Wu TT, Bilofsky H. Unusual distributions of aminoacids in complementarity-determining (hypervariable) seg-ments of heavy and light chains of immunoglobulins and theirpossible roles in specificity of antibody-combining sites. J BiolChem 1977; 252:6609–6616.

61. Padlan EA. On the nature of antibody combining sites: unusualstructural features that may confer on these sites an enhancedcapacity for binding ligands. Proteins: Struct Funct Genet1990; 7:112–124.

62. Ivanov I, Link J, Ippolito GC, Schroeder Jr HW. Constraintson the hydropathicity and sequence composition of HCDR3across evolution. In: The Antibodies. Zanetti M, Capra J,eds. Taylor & Francis; 2002:43–67.

63. Zemlin M, Klinger M, Link J, Zemlin C, Bauer K, Engler JA,Schroeder HW Jr, Kirkham PM. Expressed murine andhuman CDR-H3 intervals of equal length exhibit distinctrepertoires that differ in their amino acid composition and pre-dicted range of structures. J Mol Biol 2003; 334:733–749.

64. Lo Conte L, Chothia C, Janin J. The atomic structure of protein-protein recognition sites. J Mol Biol 1999; 285:2177–2198.

65. Collis AV, Brouwer AP, Martin AC. Analysis of the antigencombining site: correlations between length and sequencecomposition of the hypervariable loops and the nature of theantigen. J Mol Biol 2003; 325:337–354.


66. Davies DR, Cohen GH. Interactions of protein antigens withantibodies. Proc Natl Acad Sci USA 1996; 93:7–12.

67. Mian IS, Bradwell AR, Olson AJ. Structure, function and prop-erties of antibody binding sites. J Mol Biol 1991; 217:133–151.

68. Fellouse FA, Wiesmann C, Sidhu SS. Synthetic antibodiesfrom a four-amino-acid code: A dominant role for tyrosine inantigen recognition. Proc Natl Acad Sci USA 2004;101:12467–12472.

69. Folkman J. Angiogenesis in cancer, vascular, rheumatoid andother diesease. Nat Med 1995; 1:27–31.

70. Ferrara N. Role of vascular endothelial growth factor in regu-lation of physiological angiogenesis. Am J Physiol Cell Physiol2001; 280:C1358–C1366.

71. Fellouse FA, Li B, Compaan DM, Peden AA, Hymowitz SG,Sidhu SS. Molecular recognition by a binary code. J Mol Biol2005; 348:1153–1162.

72. Ashkenazi A, Dixit VM. Death receptors: signaling and modu-lation. Science 1998; 281:1305–1308.


Index

3-iodo-4-hydroxy-5-nitrophenyl-acetate (NIP), 717

a-Amylase, 597a complementation, 71

Acetyl choline receptor, 611Adjuvants, 229particulate and non-particulate,

229Affinity maturation, 497in vitro, 495in vivo, 506

Affinity-improved variants, 517Affinity-matured-binding

domains, 536Alanine scanning mutagenesis,

443Alcohol dehydrogenase, 349Alpha-helix, 196Amino acid sequence, 5, 12Amino acid, wild-type, 65, 77, 447Amino-terminal domains, 16

Angiogenesis, 294Antagonist atropine, 359Anti-ErbB2 antibody humanized

4D5, 726Anti-idiotypic antibodies, 699Antibody affinity, 685Antibody engineering, 709phage display, 709

humanization, 709Antibody library, 113Antibody library, synthetic,

709–740Antibody, murine, 495Antibody, single-heavy chains, 595Antibody-dependent cellular

cytotoxicity (ADCC), 674Antigen-binding fragments,

518, 534Antigen receptor, 623Antigen-specific B cells RNA, 533Arginine residues, 37Arginine-rich mutants, 37Assays, homogeneous, 684

741

Assays, separation-based, 684Autoimmune thrombocytopenic

purpura, 611Autologous sera, 233Automated DNA synthesis, 116Avastin2, 501Avastin Fab, and VEGF binding,

262crystallizable fragment of

(IgG-Fc), 264disulfide constraint, 264immunoglobulin G (IgG), and

Fc-III binding, 264Avidity effect, 81

B-cell blast formation, 553B-cell epitopes, 219B-lymphocyte, sensitized, 531B-lymphocyte stimulator (BLyS2),

660Barnase, 388Beta-dystroglycan, 332Binding specificity, 350Biochemical assays, 684homogeneous assays, 684separation-based assays, 684

Biotin, 145Biotin mimics, 151Biotinylation, 154casein, 145powdered milk, 145

Bone marrow, 664naıve B-cells, 664

Bovine erythrocyte carbonicanhydrase, 597

Broad-spectrum inhibitors, 284

c-erbB-2 protein, 615C-terminal domain, 534C-terminal fusions, 419Calmodulin, 477Cancer, uterine, 374Capsid, 65Capsid-encoding genes, 419Capsid proteins, minor, 20

Carboxy-terminal domain, 22Carrier proteins, 223Catalysts, 461antibodies, 463binding, 464

CD11cþ-dendritic cells, 548Cell killing, 78Cell-surface antigens, 157Cellular Braille

2

, 370Chemical mutagenesis, 124Chimeragenesis, random, 132Chimeric antibody, 495Chromo shadow domain, 335Circular dichroism, 195Clones, 389Codons, 36Codons, stop, 112Codons, TAG amber stop, 82Collagen-binding protein, 420Combinatorial infection, 126Combinatorial mutagenesis, 386Combinatorial peptide libraries,

329Competition ELISA, 543Competitive selection, 146Complementarity-determining

regions, 497, 561Complexed-antigens, 585, 586Convergent evolution, 323Copy number, 224Coregulatory proteins, 361Covalent display, 699Cre-lox recombination system,

site-specific, 126Critical-binding residues, 168Cryptic T cell epitopes, 545crystallizable fragment of

(TgG-FC), 264Cyan fluorescent protein, 292Cynomolgus BLyS, 697Cysteine residues, 44

Decahistidine tags, 401Degenerate codon, 112Degenerate homoduplex

recombination, 135

742 Index

Degenerate oligonucleotides, 112Dichroism, circular, 195Diffraction, x-ray, 26Difluoromethylphenol moiety, 480,

481Disulfide bridging, 169DNA, double-stranded, 118DNA, single-stranded, 18, 65, 118DNA shuffling, 129Double-stranded DNA, 118Double-stranded replicative form,

66Drug development, 493Drug discovery, 364DsDNA heteroduplex, 119dxr gene, 353DXR-specific peptides, 353Dystrophin, 332

E-76, and factor VII binding, 266EH domain, 322EMP1, and EPOR binding, 257erythropoietin receptor (EPOR),

and EMP1 binding, 257EMP33, and EPOR binding, 258insulinlike growth factor 1

(IGF-1), 259Enrichment ratio, 149Enzyme-linked immunosorbent

assay, 505Enzyme-mediated extension, 120Enzyme–phage particle, 469Enzyme–substrate complex, 464,

465Epitope, 442Epitope mapping, 190, 443Erbin PDZ, 270baculovirus IAP repeat (BIR),

272inhibitor of apoptosis proteins

(IAPs), 272melanoma IAP (ML-IAP), 272X-chromosome-linked IAP

(X-IAP), 272Error-prone PCR, 125

Escherichia coli, 3, 388, 721host, 120mutator strains, 124

E. coli dut–=vng– strains, 120E. coli dutþ=ungþ, 120Epitopes, TH-cell, 219Estrogen receptor, 361Estrogen receptor modulators,

selective, 361Eukaryotic cells, 566Evolution, convergent, 323Exosites, and factor VII

binding, 266

F-phage, 532F-pilus, 30, 66F1-1 ligand, 260IGF-1 binding, 260

F1-1 binding, 260helix 3 conformation, 260

Fab:CitS complex, 725Fab format, 661Fab libraries, 564Factor VIIa, 146Factor VII, exosite binding

of, 266Factor X, and factor VII binding,

267Fc-III, and IgG-Fc binding, 264Ff viruses, 3Filamentous bacteriophages, 3Filamentous phage particles, 416,

423Flanking regions, 120Flt-1D2, and VEGF binding, 262Fluorescence-activated cell

sorting, 505Folded proteins, 386Foot and mouth disease virus, 590Foreign coding sequence, 63Fos leucine zipper, 421Four-helix-bundle proteins, 387Framework library, 501Functional genomics, 416Fusion proteins, 292

Index 743

G-coupled receptors, 158G-protein coupled receptors, 355Gal4-peptide, 370, 372Gene segments, 663Genentech, Inc., 726Genotype, 534Germinal center, 665Germline gene, 672Glanzmann thrombastenia, 611Glutathione-S-transferase, 323Glycine residues, 35Growth factors, 295Guest peptide, 63Gut-associated lymphoid tissue,

558

Haptens, 166Hashimoto’s thyroiditis, 613Heavy chain enzymes, 555Helper phage genome, 419Hen eggwhite lysozyme, 202Hepatitis E virus, 620Herpes simplex virus, 222Heterohybridomas, 530Heterologous sera, 233Hexahistidine tags, 401Hexokinase, 348High-affinity S-peptide variants,

456High-throughput screening, 284Histidine tag, 399, 401Homogenous assays, 684Homolog scanning, 237Homolog shotgun scanning, 454Hormone chorionic gonadotropin,

597Horseradish peroxidase, 155Host cell membrane, 5Host genome, 416Host’s B cell pool, 540HuCAL1, 721Human growth hormone, 187Human immunodeficiency virus

type-1, 180Human kallikrein, 296

Human leukocyte antigen-DR(HLA-DR), 724

Human serum albumin, 592Humira2, 680HuPBL-SCID mice, 538Hybrid virions, 171Hybridoma technology, 203, 660synthetic libraries, 660

Hydrophobic residues, 350

IgE-antigen receptor, murine, 424IgE-binding molecules, 421IgE-binding proteins, 421, 422IgG library, species-specific, 562IGF-1 receptor (IGFR), 260IGF-binding proteins (IGFBPs),

260Immobilized tissue factor, 151Immune libraries, 531, 532Immunization, straight, 227Immunoglobulin E, 173Immunoglobulin gene cassettes,

557Immunoglobulin V genes, 547Infection, 29Infections, productive, 5Inhibitors, suicide, 467Isoprenoids, 351

Jawed vertebrates, 559Jun leucine zipper, 421

Klebsiella pneumoniae, 725Klenow fragment, 475

Leucine zipper ‘‘fastener,’’ 92Library, phage display, 63, 64, 166Liddle’s syndrome, 331Ligand elution, 152, 153Ligand, surrogate, 348Ligand–receptor interactions, 149Light chain library, 555

744 Index

Liposomes, 221loxP sites, 716Lucentis2, 510Lupus erythematosus, systemic,

609LXXLL motif, 367LXXLL peptide library, 362LymphoStat-B2, 688Lysis, T cell-mediated, 621Lytic bacteriophages, 416

M13mp series, 71Matrix metalloproteinase, 298Matrix metalloproteinase,

membrane type-1, 302Medical research council (MRC),

714Megaprimer, 121Membrane type-1 matrix

metalloproteinase, 302Metalloproteinase, matrix, 298Minor capsid proteins, 20Minus-strand origin, 66Mitochondrial membrane

transporter, 20MMP inhibitors, 302Molecular Braille�, 364Molecular recognition, 441Monkeypox virus, 605Monoclonal antibodies, 529Monoclonal islet cell antibodies,

610Monovalent display, 88, 285Morphogenesis, 65, 66Morphogenetic proteins, 11Morphosys, 720Mosaic display, 83Mouse serum albumin, 588MRC, 714MRNA population, 552Multivalent display, 285Murine antibody, 495Murine IgE-antigen

receptor, 424Mutagenesis, 386, 442

Mutagenesis, parsimonious, 686Mutagenesis, saturation, 112Mutagenesis, site-directed

scanning, 456Mutagenic cassette, 120Mutation, suppressor, 17, 18

N-terminal display, 421N-terminal domains, 557Naıve B-cells, 229Naıve human scFv repertoire, 567Naıve libraries, 660Nascent virion, 68Neisseria meningitidis, 603Neutron diffraction, 26Non-Hodgkin lymphoma, 724Nonionic detergents, 145NRIF3, 362, 363Nuclear receptors, 361

Oligo(dT) primers, 425Oligonucleotide(s), 179Oligonucleotide-directed

mutagenesis, 112, 442, 512Oligonucleotide primers, 562Oligonucleotide, random

synthetic, 531Oligonucleotide, spiked, 114Oligonucleotide synthesis, split-

pool, 449Orphan receptors, 355Osteopontin, 184

Packaging sequence, 14Panning, subtractive, 153Parsimonious mutagenesis, 686PCom3 phagemid display vector,

569PDZ domain-mediated

recognition, 269Peptide libraries on phage. See

phage-displayed peptidelibraries, 255

Index 745

Peptide(s), 220Peptides, synthetic, 284Peptides, T-specific, 358Peptide–protein pair, 354Peptide–pVIII fusions, 225Peptide-based assay, 351Peptide phage libraries, and

screening methodologies,255

Peripheral blood lymphocytes,666

Periplasmic protein, 38, 70Phage assembly, 65Phage binding, nonspecific, 145Phage binding pattern, 367Phage clones, 289Phage display, 1–3, 284, 324, 365,

390, 463, 495, 660infection, 70library technology, 165libraries, 63, 64, 166, 666packaging, 173shock protein, 90substrate, 284technology, 63, 426two-hybrid systems, 420

Phage-antibody, 541Phage-based display systems, 535Phage-based ligand selection

systems, 532Phage-display techniques, 426,

443Phage-displayed libraries, 2,

143, 500C-terminal, 270disulfide constraint, 256peptide libraries, 256, 347

Phage–host system, 2Phage library, ScFv, 160Phagemid, 72, 171, 416, 671Phagemid pHEN systems, 568Phagemid vectors, 417Phenotype, 532, 534PIII-fused Fd fragment, 564pJuFo vector, 424Plaque-forming units, 78

Plasminogen activator inhibitortype I, 294

Plasmon resonance, surface, 396Plus-strand origin, 67Poisson statistics, 41Poly(A)-priming, 417, 425Polyclonal antibodies, 529Polyclonal-pooled

immunoglobulin, 529Polymerase chain reaction, 118Polypeptide termination,

premature, 112Prion protein (PrP)-specific MAbs,

576Priretins, 22Procoat, 20Prokaryotic surface expression

systems, 416Protein design, 385Protein folding, 456Protein interaction modules, 326Protein knockout, 354Protein libraries, 442Protein–DNA link, 428Protein–ligand interactions, 423Protein–protein interactions, 336,

416extracellular, 256identification of, 255intracellular, 268

Protein-protein (PrP)-specificMAbs, 577

Proteolysis, 288, 396sticky-feet mutagenesis, 398

Proteolytic digestion, 292Proteome, 323Proton motive force, 37ProxiMol�, 678

Raman spectroscopy, 26Receptor–ligand interaction, 449Receptor–peptide complex, 370Recombinant coat protein

gene, 75

746 Index

Recombinant DNA techniques,63

Recombinant phage particle, 419Recombinant polypeptides, 111Recombinant proteins, 567Replication proteins, 6, 20Respiratory syncytial virus, 222,

603RGD motif, 713Ribosome display, 700Rolling-circle replication, 66RT-PCR-based cloning, 560

S-glycoprotein, 208S-peptide variants, high-affinity,

456Saturation mutagenesis, 112, 386Savinase, 467Scaffolds, 93, 498Scfv format, 660, 661variable heavy (VH), 660, 661variable light (VL) chains, 661

ScFv phage library, 160Screening methodologies, and

peptide phage libraries, 255Screening strategies, 143Scripps research institute, 711sequence analysis, 715

Selective estrogen receptormodulators, 361

Semen liquefaction, 296Sensitized B lymphocytes, 531Separation-based assays, 684Serum albumin, mouse, 588SHIV-macaque model, 605Shotgun scanning, 449Signal peptidane cleavage, 21Single-chain variable fragment, 127Single-heavy chains antibodies,

595Single-nucleotide point mutations,

513Single-stranded DNA, 18, 65, 118Site-directed scanning

mutagenesis, 456

Site-specific Cre-lox recombinationsystem, 126

Somatic hypermutation, 665Species-specific IgG library, 562Spiked oligonucleotide, 114Split-pool oligonucleotide

synthesis, 449Sticky feet mutagenesis, 386Stoffel fragment, 475Stop codons, 112Straight immunization, 227Streptavidin, 158europium cryptate, 370

Substrate phage display, 284Subtiligase, 474Subtilisin, 467Substrate phage library,

subtracted, 293Subtracted substrate phage

library, 293Subtractive panning, 153Suicide inhibitors, 467Suppressor mutations, 17, 18Surface plasmon resonance, 396Surrogate ligands, 348SwissProt, 298Synthetic antibody libraries,

709–740advantages, 710antibody frameworks, 721antigen-binding activity, 719anti-murine vascular

endothelial growth factor(mVEGF), 728

Synthetic peptides, 284Systemic lupus erythematosus,

609autonomous VH domains, 720binding affinities, 717camelid antibodies, 720complementarity determining

region (CDR), 711

T cell-mediated lysis, 621T-specific peptides, 358

Index 747

TH-cell epitopes, 219TAG amber stop codons, 82Taq DNA polymerase, 125Tetanus toxoid, 201Therapeutic antibodies, 659Thioredoxin, 25Thrombin cleavage site, 148Thrombocytopenic purpura,

autoimmune, 611Thyroid hormone receptor binding

protein, 362TMB8, 360tolQ and tolR genes, 30Trabio2, 686Transducing units, 78Transition state, 464Transition-state analogues, 463Trypsin, 476Two plasmid system, 125Two-hybrid screens, 328Type 3 ‘‘fusion phage,’’ 77Type n systems, 75

Ubiquitin, 390Uracil-DNA glycosylase, 133Uterine cancer, 374Utrophin, 332

V genes, 533, 562, 716v107 peptide class, and VEGF

binding, 262v108 peptide class, and VEGF

binding, 262Vaccinia virus, 605Vascular endothelial growth factor

(VEGF), 261, 295tyrosine kinase receptors of, 261

Vascularization, 295VEGF binding, and Avastin Fab,

262Viral elongation, 25Viral morphogenesis, 39Virion assembly, 173VP16-receptor, 370, 372

Walker nucleotide binding motif,25

Western blot analysis, 428Whole-Agn libraries, 173Wild-type amino acid, 65, 77, 447

X-ray diffraction, 26

748 Index

2005-[sachdev s. sidhu] phage display in biotechnology

Documents